Safeguard
Supply Chain Security•

How Bluetooth Gets Hacked and How to Secure It

Bluetooth security: how phones, earbuds, wearables and cars get attacked over Bluetooth, from BlueBorne to BLUFFS, and how to harden every layer.

Safeguard Security Team
61 min read

Bluetooth gets hacked in three main ways. Attackers abuse weak or downgraded pairing to impersonate a trusted device or decrypt traffic. They send malformed packets that crash or take over a buggy Bluetooth stack or chip firmware. And they listen to what devices broadcast, to track people or to relay a phone's presence and unlock a car or door. Almost every serious case needs the attacker within radio range, and almost every one is fixed by updates, stronger pairing settings and better firmware from the companies that build the devices.

That is the short answer. This guide is the long one. It explains how Bluetooth works on phones, laptops, headphones, wearables and cars, what software runs it, and where that software can be attacked. It walks through the vulnerability families that defined the last decade, from BlueBorne in 2017 to BLUFFS in 2023 and the headphone and car stack flaws of 2025, with the CVE numbers, what each attack needs, what it lets an attacker do, how to notice it and how to defend against it. It covers tracking through Bluetooth Low Energy (BLE) advertisements, the abuse of item trackers and the anti-stalking features built in response, and the relay problem behind phone-as-a-key car access. It ends with hardening checklists for home users, businesses and device makers, and with the part most guides skip: the chipset SDKs and firmware that one vendor writes and thousands of products ship.

It is written from the attacker's point of view for defenders. You will find how each attack works and why, but no exploit code, no tool command lines and no step-by-step instructions. Where a tool is named, it is named as context for authorised testing.

A note on the law. Testing Bluetooth devices you do not own, or have no written permission to test, is illegal in most countries. Everything in this guide is for protecting your own devices, your organisation's devices with authorisation, or the products you build.

This post is part of a device security series. The companion guide on how Wi-Fi gets hacked and how to secure it covers the other radio in your pocket, and our teardown of how a laptop is built and the post on fabless chip designers and the software supply chain explain where Bluetooth chips and their firmware come from.

Contents

  1. How does Bluetooth work, and what software runs it?
  2. How does Bluetooth pairing work, and which pairing modes are weak?
  3. What is the Bluetooth attack surface?
  4. What does an attacker need to hack Bluetooth?
  5. Which Bluetooth vulnerabilities matter most?
  6. How are people tracked through BLE advertisements?
  7. How are Bluetooth trackers abused, and how do anti-stalking features work?
  8. How secure are Bluetooth car keys and phone-as-a-key systems?
  9. How do you detect a Bluetooth attack?
  10. Why do Bluetooth flaws spread through the supply chain?
  11. How should home users secure Bluetooth?
  12. How should businesses secure Bluetooth?
  13. How should device makers secure Bluetooth?
  14. How Safeguard helps teams that build Bluetooth products
  15. The takeaway
  16. Frequently asked questions
  17. Sources

Diagram of the Bluetooth attack surface: phones, laptops, cars, headphones, wearables and trackers connect to eight layers from discovery and pairing to firmware updates and companion apps

Figure 1. The layers of Bluetooth that an attacker within radio range can reach, across the devices people carry.

How does Bluetooth work, and what software runs it?

Bluetooth is two radio technologies under one name. Bluetooth Classic (BR/EDR) carries audio, calls and keyboards. Bluetooth Low Energy (BLE) carries small bursts of data for wearables, trackers, locks and sensors. Each device splits the work between a host stack in the operating system and controller firmware on the Bluetooth chip.

Two terms do most of the work in this guide, so here they are in plain language.

Bluetooth Classic, formally Basic Rate and Enhanced Data Rate (BR/EDR), is the older and higher-throughput form. It is what your earbuds use for music (the A2DP profile), what your car uses for hands-free calls (HFP), and what a Bluetooth keyboard uses to type (HID, the human interface device profile).

Bluetooth Low Energy (BLE) arrived with version 4.0 of the specification. It is built for small, infrequent messages and long battery life. A fitness band, an item tracker, a smart lock or a glucose monitor almost always talks BLE. BLE devices announce themselves with short broadcast packets called advertisements, and they expose data and controls through the Generic Attribute Profile (GATT), a simple model of services and readable or writable values called characteristics.

Many products are dual mode: a phone, a laptop or a pair of premium earbuds speaks both. That matters for security because some attacks cross from one transport to the other, as you will see with BLURtooth.

The host and the controller

Every Bluetooth device runs two very different pieces of software.

The controller is the Bluetooth chip and its firmware. It handles the radio, the link layer (for BLE) or link manager (for Classic), the encryption of packets on the air and the negotiation of key sizes. Controller firmware usually comes from the chipset vendor's software development kit (SDK) and is almost always closed source. When a packet arrives over the air, the controller is the first code to parse it.

The host is the Bluetooth stack in the operating system. It implements the protocols above the link: L2CAP (the multiplexing layer), SDP (service discovery), RFCOMM (serial emulation), BNEP (networking), ATT and GATT (for BLE data), and SMP, the Security Manager Protocol that runs BLE pairing. On Android, the default stack ships with the operating system and has supported BLE since Android 4.3, according to the Android Open Source Project. Linux uses BlueZ. Apple and Microsoft ship their own stacks. Embedded devices often use an open source stack such as Zephyr's or a stack from the chip vendor or a third party.

The two halves talk over the host controller interface (HCI), a defined set of commands and events that runs over UART, USB, SDIO or shared memory.

Layered diagram of the Bluetooth software stack: apps, profiles, host stack, host controller interface, controller firmware and radio, with who supplies each layer

Figure 2. The Bluetooth software stack and the companies that typically supply each layer.

On a phone or laptop the host stack is part of the operating system and gets security updates with it. On earbuds, trackers and many wearables, the host stack and the controller firmware often run on a single system on a chip (SoC) and come from one SDK. That design choice is why a single SDK flaw can sit inside hundreds of brands of earbuds, a theme this guide returns to.

Where Bluetooth lives in each device type

  • Phones run full dual mode stacks and talk to everything: earbuds, watches, cars, trackers, keyboards. They also act as the scanning network that finds lost item trackers.
  • Laptops carry a combined Wi-Fi and Bluetooth module, with controller firmware loaded by the operating system driver. Our laptop teardown covers where those modules and their firmware come from.
  • Headphones and earbuds run a Bluetooth audio SoC with the vendor's SDK and a thin layer of brand-specific code, plus a companion app on the phone.
  • Wearables such as watches and fitness bands run BLE, an embedded operating system and a companion app, and often sync health data to the cloud.
  • Cars run Bluetooth in the infotainment head unit for calls, music and contacts, and increasingly in a separate access module for phone-as-a-key entry.

A short history of Bluetooth security features

The security model changed several times, and older modes are still in the field.

  • Before version 2.1, pairing used a PIN. Headsets with no keypad shipped fixed PINs such as 0000 or 1234, as Wikipedia's summary of the specification notes.
  • Version 2.1, adopted in July 2007, introduced Secure Simple Pairing (SSP), which uses public-key cryptography.
  • Version 4.0 introduced BLE with its own pairing, now called LE legacy pairing, and BLE privacy with randomised addresses.
  • Version 4.1 added Secure Connections to Bluetooth Classic, and version 4.2 (December 2014) brought LE Secure Connections, which uses elliptic curve Diffie-Hellman (ECDH) on the P-256 curve, according to Apple Platform Security and Wikipedia.
  • Version 5.4 (February 2023) added Encrypted Advertising Data, and version 6.0 (August 2024) added Channel Sounding, a way to measure distance between devices, both covered later.

How does Bluetooth pairing work, and which pairing modes are weak?

Pairing is how two devices create shared secret keys; bonding is storing them for later. The weak modes are legacy PIN pairing, LE legacy pairing and Just Works, which gives no protection against a man in the middle. Numeric Comparison, Passkey Entry and Secure Connections Only mode are the strong choices.

When you pair earbuds with a phone, the two devices run a key agreement and store the result. Apple's platform security guide describes the pieces cleanly: pairing creates shared keys, bonding stores them for future connections, authentication checks both sides hold the same keys, and encryption protects the traffic.

The pairing methods

Secure Simple Pairing and LE Secure Connections both use ECDH, so a passive listener cannot work out the key. What differs between methods is protection against a man in the middle (MITM), an attacker who sits between the two devices and relays or alters what they say.

  • Just Works. No code, no comparison. It exists for headsets and sensors with no screen or buttons. It protects against passive eavesdropping when used with SSP or LE Secure Connections, but it provides no MITM protection.
  • Numeric Comparison. Both devices show a six digit number and the user confirms they match. It protects against MITM only if the user actually compares the numbers.
  • Passkey Entry. One device shows a six digit code and the user types it on the other, or the user types the same code on both. It protects against MITM.
  • Out of Band (OOB). Some pairing data travels over another channel, usually NFC. It is as strong as that other channel.

The device's input and output capabilities decide which method gets used. A pair of earbuds with no screen will fall back to Just Works. That fallback is the single most common weakness in consumer Bluetooth.

Legacy modes that are still out there

Legacy PIN pairing derives keys from a short PIN. A fixed PIN printed in a manual is not a secret.

LE legacy pairing, used by BLE devices before LE Secure Connections, uses a short temporary key that can be brute forced from a captured pairing exchange. Mike Ryan showed at USENIX WOOT 2013 that the encryption of BLE links using this pairing could be rendered useless. Many early wearables and sensors used it, and some cheap devices still do.

Key sizes and Secure Connections Only mode

Bluetooth Classic negotiates an encryption key length between 1 and 16 bytes. That negotiation is what the KNOB attack abused. Since then the Bluetooth Special Interest Group (SIG), the body that maintains the specification, recommends rejecting links with keys shorter than 7 octets, and for implementations that always use Security Mode 4 Level 4, rejecting anything below 16 octets, according to its BLUFFS security notice.

Secure Connections Only mode is a setting in which a device refuses any connection that does not use Secure Connections. The same SIG notice says that having both devices in this mode ensures sufficient key strength, and CERT/CC notes that it blocks the downgrade step in the BIAS attack. It is the closest thing Bluetooth has to a "strict mode", and every device maker should know whether their product can run in it.

Table comparing Bluetooth pairing methods by protection against eavesdropping and man in the middle attacks, with typical uses

Figure 3. Bluetooth pairing methods compared. Just Works and legacy modes are where most practical attacks start.

What is the Bluetooth attack surface?

The Bluetooth attack surface is every piece of code that handles radio input before and after pairing: advertisements, the pairing exchange, controller firmware, host stack parsers, profiles and GATT services, vendor debug protocols, the firmware update path, and the companion apps and cloud accounts behind the device.

Think about it from the attacker's side. They are within radio range and can send any packet they like, including packets no compliant device would ever send. What will parse it?

  1. Discovery and advertising. BLE devices broadcast advertisements to anyone listening. Classic devices respond to inquiries when discoverable, and as ERNW notes in its BlueFrag write-up, many accept connections when addressed even if they are not discoverable. This layer leaks identity and is the entry point for tracking.
  2. Pairing and key exchange. Weak methods, downgrades and key size negotiation live here. KNOB, BIAS, BLURtooth and BLUFFS all target this layer.
  3. Controller firmware. The chip's link layer and link manager parse packets first, often before any authentication. SweynTooth and BrakTooth are flaws in this layer.
  4. Host stack. L2CAP, SDP, BNEP and SMP parsers run in the operating system kernel or a privileged daemon. BlueBorne, BlueFrag and BleedingTooth are flaws here.
  5. Profiles and GATT services. Audio, calling, keyboard input, health data, lock control and car keys. A GATT characteristic that allows writes without authentication is a door left open. CVE-2023-45866 abused the HID profile.
  6. Vendor custom protocols. Chip and device makers add their own channels for tuning, testing and updates. The Airoha headphone flaws of 2025 sat in one of these.
  7. Firmware update path. Over the air (OTA) updates for earbuds, watches and cars. If updates are unsigned or unauthenticated, the update channel becomes an attack channel.
  8. Companion apps and cloud. The phone app that configures the device, the account it registers to, and location networks such as Apple's Find My or Google's Find Hub.

Most of the severe public vulnerabilities in the last ten years needed no pairing at all. They hit code that runs before authentication: controller firmware, the host stack's early parsers, or a service that never asked for authentication.

What does an attacker need to hack Bluetooth?

Most Bluetooth attacks need physical proximity, inexpensive hardware and either a known flaw on an unpatched device or a weak configuration. Some need only the target's Bluetooth address. Others need the attacker near two devices at the moment they connect. Almost none can be launched from across the internet.

The requirements vary by attack class, and they matter for risk ranking.

  • Radio range. Bluetooth is short range. PCA Cyber Security describes it as usually within about 10 metres in its PerfektBlue explainer. An attacker has to be nearby, or have a device nearby. That rules out mass remote exploitation but makes airports, offices, trains, car parks and conferences the natural venues.
  • Commodity hardware. The BrakTooth researchers needed only a low-cost ESP32 development kit with modified link manager firmware and a PC. The researcher who found CVE-2023-45866 reported that the keystroke injection could be launched from a Linux computer with a standard Bluetooth adapter.
  • A target address. For Classic attacks, the attacker usually needs the victim's Bluetooth device address. ERNW notes that for BlueFrag, the address was all the attacker needed to know.
  • Timing, for protocol attacks. KNOB, BIAS and BLUFFS need the attacker in range while two specific devices connect or reconnect, and some need precise radio timing. That makes them specialist attacks.
  • An unpatched device or a weak configuration. Nearly every case in this guide has a fix. The attack works because the fix has not arrived, which for embedded devices can mean never.

On the authorised testing side, researchers have published tools and proofs of concept for many of these families, such as the BrakTooth proof of concept, which the ASSET Research Group made available to vendors to validate their stacks, and the BLUFFS toolkit on GitHub. Security teams and product makers use such tools in labs, on their own hardware, under written authorisation. This guide does not describe how to run them.

Which Bluetooth vulnerabilities matter most?

The most important Bluetooth vulnerability families are BlueBorne (2017), KNOB (2019), SweynTooth, BIAS and BLURtooth (2020), BrakTooth (2021) and BLUFFS (2023), plus implementation flaws like BlueFrag, BleedingTooth, CVE-2023-45866, the 2025 Airoha headphone flaws and PerfektBlue in car head units.

These fall into two groups, and the difference changes who has to fix them.

  • Specification flaws live in the Bluetooth Core Specification itself. Any device that follows the standard faithfully is exposed. KNOB, BIAS, BLURtooth and BLUFFS are in this group, as is the invalid curve weakness of 2018. Fixing them requires a specification change and then updates from every controller and host vendor.
  • Implementation flaws are bugs in a particular stack or chip SDK. BlueBorne, BlueFrag, BleedingTooth, SweynTooth, BrakTooth, the keystroke injection bugs, the Airoha flaws and PerfektBlue are in this group. They need a fix from whoever wrote the code, and then from everyone who ships it.

Timeline of major Bluetooth vulnerability families from BlueBorne in 2017 to PerfektBlue in 2025, marked as specification or implementation flaws

Figure 4. Major Bluetooth vulnerability families, 2017 to 2025. Purple marks specification flaws, red marks implementation flaws.

BlueBorne (2017): remote code execution in four operating systems

How it worked. In September 2017, the security firm Armis disclosed a set of eight vulnerabilities in the Bluetooth stacks of Android, Linux, Windows and older iOS, which it named BlueBorne. CERT/CC note VU#240311 lists them:

  • CVE-2017-1000251, a stack buffer overflow in the Linux kernel's L2CAP handling (BlueZ), from kernel 3.3-rc1 onward;
  • CVE-2017-1000250, an out-of-bounds read in BlueZ's SDP server that leaks memory;
  • CVE-2017-0785, a separate SDP information leak in Android;
  • CVE-2017-0781 and CVE-2017-0782, a heap overflow and an integer underflow in Android's BNEP handling, both leading to remote code execution;
  • CVE-2017-0783 and CVE-2017-8628, logic flaws in the PAN profile on Android and Windows that enable man-in-the-middle attacks;
  • CVE-2017-14315, a heap overflow in Apple's Low Energy Audio Protocol on iOS 9.3.5 and earlier.

Why it worked. These were parsing bugs and logic errors in code that runs before or during connection setup. CERT/CC describes the attacker as unauthenticated and remote within radio range.

Impact. Information disclosure and arbitrary code execution, with a CVSS base score of 7.9 in the CERT/CC note. Armis estimated at the time that more than 8.2 billion devices had Bluetooth and that 5.3 billion were at risk, and a year later estimated that over 2 billion were still exposed, as summarised on Wikipedia.

Detection. On the device, a successful exploit looks like nothing, or like a Bluetooth service crash. At the fleet level, the useful signal is the patch level: any Android device below the September 2017 security patch level, and any Linux system with an unpatched kernel and BlueZ, was exposed.

Defence. Patches shipped for Windows, iOS, the Linux kernel and Android. CERT/CC pointed out that some Android phones would get delayed updates or none, and recommended disabling Bluetooth where no update was available. BlueBorne is the clearest early lesson that "Bluetooth on" is network exposure, and that the slowest link in the patch chain is usually the device maker.

The invalid curve attack (2018): CVE-2018-5383

How it worked. Secure Simple Pairing and LE Secure Connections use ECDH. Both sides exchange public keys on an agreed elliptic curve. Some implementations did not check that the received public key was actually a valid point on that curve. CERT/CC note VU#304725 explains that a remote attacker within range could inject an invalid public key and determine the session key with high probability.

Impact. A man in the middle could passively decrypt traffic or forge and inject messages. Both BLE Secure Connections in operating system software and Classic SSP in device firmware could be affected.

Defence. The Bluetooth SIG updated the specification to require validation of any public key received during pairing and added a test for it to its qualification program. For device makers, this is the template lesson for cryptography in embedded stacks: validate every input to a key exchange, even if the standard once left it implicit.

KNOB (2019): forcing a one-byte encryption key

How it worked. The Key Negotiation of Bluetooth (KNOB) attack, by Daniele Antonioli, Nils Ole Tippenhauer and Kasper Rasmussen, was presented at USENIX Security 2019 and tracked as CVE-2019-9506. Bluetooth Classic lets two devices negotiate how many bytes of entropy their encryption key has, anywhere from 1 to 16. That negotiation was not integrity protected. CERT/CC's note VU#918987 walks through it: an attacker intercepts the proposal and lowers it to one byte, both devices accept because one byte is within the allowed range, and both report encryption as active.

What the attacker needs. Radio range of two devices as they set up an encrypted connection, and the ability to inject packets at the right moment.

Impact. With one byte of entropy there are only 256 possible keys, so the attacker can brute force the key and decrypt the session. The researchers tested 24 devices built on more than 17 chips from vendors including Broadcom, Qualcomm, Apple and Intel, and all were vulnerable, according to the KNOB site.

Detection. Users cannot see a key length. Device makers and testers can, in controller logs and over-the-air captures in a lab.

Defence. The Bluetooth SIG issued Expedited Errata Correction 11838, and CERT/CC directed host and controller suppliers to it, with downstream vendors told to get updates from their suppliers. The SIG's current guidance, restated in its BLUFFS notice, is a minimum key length of 7 octets, and 16 octets for devices that can always use the strongest security mode. The KNOB site warns that devices not updated since late 2018 are likely still vulnerable.

SweynTooth (2020): crashing and bypassing BLE chip SDKs

How it worked. Matheus Garbelini, Sudipta Chattopadhyay and Chundong Wang of the Singapore University of Technology and Design fuzzed the BLE stacks in system on chip SDKs. Their disclosure released 12 vulnerabilities on 11 February 2020 across SDKs from seven SoC vendors: Texas Instruments, NXP, Cypress, Dialog Semiconductor, Microchip, STMicroelectronics and Telink. A second batch of five followed on 14 July 2020, affecting stacks from Texas Instruments, Espressif, Microchip, ON Semiconductor and the Zephyr project.

Why it worked. The flaws were in the link layer and early protocol handling, which runs before pairing. Malformed packets such as oversized length fields, truncated L2CAP frames or invalid connection requests were not handled safely.

Impact. The researchers grouped them into three types:

  • Crash: memory corruption or a hard fault that restarts the device, if the product handles restarts at all;
  • Deadlock: the BLE connection hangs until someone power cycles the device;
  • Security bypass: the worst type. The Zero LTK Installation flaw (CVE-2019-19194) let an attacker bypass Secure Connections pairing and gain read and write access to functions meant only for authorised users.

The researchers noted that a search of the Bluetooth listing site for products using the affected SoCs returned about 480 listings, each possibly covering several products, and that they had identified medical and logistics products that could be affected.

Detection. Unexplained reboots or hangs in BLE devices are the visible symptom. Device makers can test their own builds against the published cases.

Defence. Almost all SoC vendors released patches during disclosure. The researchers' key point is the one this guide is built around: IoT products relying on those SoCs still needed separate patches from their own vendors, and only if those vendors supported firmware updates at all.

BIAS (2020): impersonating a device you already trust

How it worked. The Bluetooth Impersonation Attacks (BIAS), CVE-2020-10135, were presented at the IEEE Symposium on Security and Privacy in May 2020 by the KNOB team. CERT/CC note VU#647177 explains the mechanics at a conceptual level. When two previously paired devices reconnect, the attacker spoofs the address of one of them. If the devices originally paired with Secure Connections, the attacker claims not to support it, downgrading to legacy authentication, unless the target is in Secure Connections Only mode. The attacker then uses a role switch to become the side that authenticates the other, and legacy authentication does not require mutual authentication, so the target completes authentication without the attacker ever knowing the link key.

What the attacker needs. The address of a device the victim has bonded with, and radio range of the victim. The real paired device does not need to be present.

Impact. Full access as the trusted device. The researchers tested 30 devices with 28 different Bluetooth chips from Cypress, Qualcomm, Apple, Intel, Samsung and CSR, and all were vulnerable, according to the BIAS site. BIAS can be combined with KNOB.

Defence. The SIG published a statement on 18 May 2020, and vendors shipped controller and host updates. Secure Connections Only mode blocks the downgrade. The BIAS site's warning mirrors KNOB's: devices not updated after December 2019 are likely vulnerable.

BlueFrag and BleedingTooth (2020): host stack memory bugs

Two 2020 cases show that the host stack in the operating system is a target in its own right.

BlueFrag, CVE-2020-0022, was found by Jan Ruge of ERNW. On Android 8.0 to 9.0 it allowed remote code execution with no user interaction, provided Bluetooth was on and the attacker knew the device's Bluetooth address. ERNW's write-up notes that the underlying flaw was still present in Android 10 but harder to exploit there, and that it was fixed in the February 2020 Android security patch.

BleedingTooth was a set of Linux kernel Bluetooth flaws published by Google's security research team in October 2020. The BadKarma advisory, CVE-2020-12351, describes a heap-based type confusion in the kernel's L2CAP code affecting Linux 4.8 and later. A remote attacker at short distance who knew the victim's address could cause a denial of service or possibly execute code with kernel privileges. The advisory adds that a malicious Bluetooth chip could trigger it too, a reminder that the host must not blindly trust its own controller.

Defence. Operating system updates. For Linux-based products, that means tracking kernel and BlueZ versions in every firmware image, which is a software composition problem as much as a patching one.

BLESA (2020): spoofing BLE reconnections

Researchers from Purdue University and EPFL presented BLESA at USENIX WOOT 2020, where it won best paper. Their formal analysis of how two previously connected BLE devices reconnect found two weaknesses in the specification, so that even a correct implementation could be vulnerable to spoofing. BLESA let an attacker impersonate a BLE device and feed spoofed data to a device it had previously paired with. The researchers found it could be carried out easily against some implementations, including Linux's, and found a logic bug enabling it in the Android and iOS stacks, which Google and Apple acknowledged.

The impact is clearest for sensors. A phone that trusts a paired glucose monitor or fitness band could be fed false readings. The defence is making sure reconnection enforces the same authentication and encryption as the original bond, and failing closed when it does not.

BLURtooth (2020): overwriting keys across transports

How it worked. Cross-Transport Key Derivation (CTKD), introduced in Bluetooth 4.2, lets a dual mode device pair once over Classic or BLE and derive keys for the other transport too. The HexHive group at EPFL, with researchers from CISPA and Oxford, found that an attacker could use CTKD to overwrite a stronger, authenticated key with a weaker or unauthenticated one, tracked as CVE-2020-15802.

What the attacker needs. According to the Bluetooth SIG notice, a device in range that supports both transports and CTKD and that permits pairing on either transport with no authentication, such as Just Works, or with no user-controlled restriction on when pairing is allowed.

Impact. Impersonation and man-in-the-middle attacks between devices that had bonded with authenticated pairing. CERT/CC note VU#589825 explains that an attacker could gain access to profiles or services that should have been restricted.

Defence and a disagreement. The SIG says the issue affects implementations of specifications 4.2 through 5.0 and recommends adopting the CTKD restrictions mandated in version 5.1 and later, which forbid an overwrite that reduces key strength or MITM protection. The researchers state that they tested Bluetooth 5.1 and 5.2 devices and still found them vulnerable, and that the SIG published its notice without informing them. Their paper was later published at AsiaCCS 2022. For defenders, the safe reading is to treat the version number as no guarantee, require authenticated pairing for anything sensitive, and limit how long a device stays in pairing mode.

BrakTooth (2021): crashing Classic chip firmware

How it worked. The ASSET team behind SweynTooth turned to Bluetooth Classic. BrakTooth, disclosed at the end of August 2021, is a family of 16 vulnerabilities in the link manager and baseband firmware of commercial Classic stacks. The team evaluated 13 devices from 11 vendors, including an Intel AX200 in a laptop, a Qualcomm chip in a phone, Espressif's ESP32, Cypress, Texas Instruments and several audio SoCs used in speakers and headphones. At the time of disclosure 25 CVEs had been assigned.

What the attacker needs. A cheap ESP32 board with non-compliant link manager firmware and a PC, in range of the target.

Impact. Mostly denial of service through firmware crashes and deadlocks, but the Feature Pages Execution flaw on the ESP32, CVE-2021-28139, allowed arbitrary code execution. The researchers found that BrakTooth affected over 1,400 product listings and warned that because the Bluetooth stack is often shared across many products, many more were probably affected. Later updates added MediaTek patches (CVE-2022-20021 to 20023) and noted that some MacBooks and iPhones were affected as well.

Defence. Firmware updates from chip vendors, passed through to products. The disclosure also records the delay this creates: some vendors with CVE authority issue CVEs only after a patch is distributed, so for weeks after disclosure some flaws had no tracker at all.

BLUFFS (2023): breaking forward and future secrecy

How it worked. Daniele Antonioli, now at EURECOM, published six attacks named BLUFFS (Bluetooth Forward and Future Secrecy), tracked as CVE-2023-24023, in 2023. According to BleepingComputer's summary, the attacks exploit four flaws in how Classic session keys are derived, two of them new. The Bluetooth SIG notice explains the core idea: an attacker spoofing two bonded devices to each other can push them into legacy encryption establishment, force the minimum key length both support, and fix the values that should randomise the key. The same weak session key is then used for every session while the attacker is nearby.

Impact. If the key can be brute forced, every past and future session the attacker recorded can be decrypted, which is what forward and future secrecy are meant to prevent. With keys below 7 octets, brute forcing in real time allows live injection. The SIG says any conforming Classic implementation is expected to be vulnerable, across Core Specifications 4.2 through 5.4. The researchers tested phones, earphones and laptops running Bluetooth 4.1 to 5.2, and all were susceptible to at least three of the six attacks.

Defence. The SIG recommends rejecting service-level connections with keys below 7 octets, requiring 16 octets where Security Mode 4 Level 4 is always possible, running Secure Connections Only mode, and remembering that a bond was created with Secure Connections so that any later downgrade terminates the link. The researchers also proposed a hardened key derivation function. The BLUFFS repository holds a checker that vendors can use on their own devices.

CVE-2023-45866 (2023): a fake keyboard that pairs itself

How it worked. The researcher behind the 2016 MouseJack work on wireless keyboards found that several Bluetooth stacks could be tricked into pairing with a fake keyboard without user confirmation, then accept keystrokes from it. The write-up describes it as a mix of an unauthenticated pairing mechanism defined in the HID specification and implementation bugs that exposed it to an attacker.

What the attacker needs. Radio range and a Linux computer with a standard Bluetooth adapter. Conditions differed by platform: unpatched Android devices were vulnerable whenever Bluetooth was on; Linux with BlueZ when Bluetooth was discoverable and connectable; iOS and macOS when Bluetooth was on and a Magic Keyboard had been paired.

Impact. Typing as the user. The attacker could do anything that did not require a password or biometric. The researcher reproduced it on Android back to version 4.2.2 from 2012. For Linux the write-up records a sharp lesson: a 2020 fix (CVE-2020-0556) had shipped disabled by default, and the 2023 BlueZ patch turned it on.

Detection. A new keyboard appearing in Bluetooth settings that nobody paired, or text and actions appearing on their own. In managed fleets, an alert on any newly bonded HID device is a high-value control.

Defence. Operating system updates from Google, the Linux distributions and Apple, and a policy that keyboards and other input devices may only pair after explicit user action.

Airoha headphone and earbud SoCs (2025): an open debug channel

How it worked. In June 2025, Dennis Heinze and Frieder Steinmetz of ERNW published an advisory on Bluetooth headphones and earbuds built on Airoha SoCs. Airoha supplies chips, reference designs and an SDK used widely in true wireless earbuds. The devices exposed a powerful custom protocol that could read and write the device's RAM and flash. It was reachable over BLE GATT by an unpaired attacker and over a Classic RFCOMM channel without pairing. Three CVEs cover it: CVE-2025-20700 (missing authentication for GATT services), CVE-2025-20701 (missing authentication for Bluetooth Classic) and CVE-2025-20702 (critical capabilities of a custom protocol).

What the attacker needs. ERNW is explicit: being in Bluetooth range is the only precondition. No pairing, no user interaction.

Impact. In most cases, full takeover of the headphones. ERNW demonstrated reading what was playing from memory and listening through the microphone by opening a hands-free connection. Because the attacker can read the device's memory, they could also take over the trust relationship the earbuds have with the paired phone. The researchers confirmed vulnerable models from Beyerdynamic, Bose, Jabra, JBL, Marshall, Sony and others, and noted that some brands did not know they were using an Airoha SoC because they had outsourced the Bluetooth module.

Defence. Firmware updates from the headphone makers, once they received the fixed SDK. For buyers, keeping the companion app current. For device makers, the lesson is blunt: debug and tuning protocols must be removed or locked behind authentication in production builds, and you must know which SoC and SDK your module supplier used.

PerfektBlue (2025): a third-party stack inside cars

How it worked. PCA Cyber Security found four vulnerabilities in OpenSynergy's BlueSDK, a closed source Bluetooth stack widely used in automotive infotainment. Chained together they gave remote code execution. The PerfektBlue advisory, public on 7 July 2025, lists CVE-2024-45434 (a use-after-free in the AVRCP service, the most severe), CVE-2024-45431 (improper validation of an L2CAP channel identifier), CVE-2024-45433 (incorrect function termination in RFCOMM) and CVE-2024-45432 (a function call with incorrect parameters).

What the attacker needs. Pairing with the target, which depending on how each carmaker configured the stack could take at most one click from the user, according to the PerfektBlue site.

Impact. Code execution on the head unit. From there, PCA says, an attacker could track GPS coordinates, record audio in the car, read the phonebook and attempt lateral movement toward other electronic control units. PCA demonstrated it on head units in Volkswagen, Mercedes-Benz and Skoda vehicles.

The supply chain detail that matters. PCA reported the flaws in May 2024. OpenSynergy rolled out patches to its customers in September 2024. Yet PCA's timeline records that not all carmakers had received the patch by June 2025, and that one carmaker told PCA it had not received the notice or patch from its supply chain at all. PCA put this down to long and complex vehicle supply chains. The fix existed for nine months before some vehicle makers had it.

Defence. Owners should install vehicle software updates, ask the dealer whether their head unit is affected, and turn off Bluetooth in the car if no update is available. Carmakers and their suppliers need an inventory that answers "which of our head units run BlueSDK, and which version" in hours, not months.

Table of Bluetooth vulnerability families showing transport, layer, whether pairing was needed, and who ships the fix

Figure 5. Where each vulnerability family lives and who has to ship the fix. Specification flaws need every vendor to change; chip flaws need the SDK fix to reach each product.

How to weigh these risks

Not every family deserves the same urgency. Protocol attacks such as KNOB, BIAS and BLUFFS are clever and broad, but they need an attacker near two specific devices at the right moment. Unauthenticated access to a debug protocol, or a stack that pairs a keyboard silently, needs only range and a laptop. Figure 6 is our qualitative reading of the public research, intended to help teams decide what to fix first.

Two by two matrix of Bluetooth attack classes by attacker effort and impact, from fix first to specialist attacks

Figure 6. Risk versus attacker effort for common Bluetooth attack classes. This is a qualitative reading of public research, not a scoring standard.

How are people tracked through BLE advertisements?

BLE devices broadcast advertisements that anyone nearby can receive. If the address or payload stays stable, scanners can link sightings and follow a person. BLE privacy rotates the address, but research shows stable payload fields can still carry identity across rotations, so makers must rotate everything together.

Every BLE peripheral that wants to be found broadcasts advertisements several times a second: an address, some flags, perhaps a name, service identifiers and manufacturer data. Nothing about that is secret. A cheap receiver in a shop, a station or an office lobby can log every device that walks past.

What BLE privacy does

The Bluetooth SIG explains in its blog on privacy that LE Privacy replaces the device's real address in advertisements with a random value that changes at an interval the manufacturer sets. Devices that have paired exchange an Identity Resolving Key (IRK), which lets a trusted device recognise the random addresses. To everyone else, one device looks like a series of unrelated devices. These features have existed since Bluetooth 4.0, and version 4.2 added link layer privacy. Apple describes the same mechanism, which it calls address randomisation, in its platform security guide.

Why randomised addresses are not enough

In Tracking Anonymized Bluetooth Devices, presented at PETS 2019, Johannes Becker, David Li and David Starobinski of Boston University showed that many devices using address randomisation could still be tracked passively well beyond their rotation cycles. The trick was identifying tokens in the advertising payload. Because the payload and the address did not change at exactly the same moment, a scanner could carry identity from the old address to the new one. They also found an identity-exposing attack through a device accessory and a side channel on iOS, and they proposed countermeasures.

Other leaks come from device names that include a person's name, long rotation intervals, and accessories that advertise constantly with fixed identifiers.

What helps

  • For device makers: rotate the address and every identifying payload field at the same moment; avoid counters and fixed vendor bytes; stop advertising when not needed; use Encrypted Advertising Data, added in Bluetooth 5.4, for payload bytes that only intended receivers should read.
  • For users: do not put your name in device names, turn off Bluetooth on accessories you are not using, and remove old pairings.
  • For businesses: treat BLE beacons and analytics sensors as personal data processing where they log visitor devices.

Diagram of what a BLE device broadcasts and how address randomisation works, beside the unwanted tracker rules for near-owner and separated modes

Figure 7. Left: what BLE advertisements expose and how privacy features help. Right: the anti-stalking rules for item trackers in the IETF draft.

How are Bluetooth trackers abused, and how do anti-stalking features work?

Item trackers are BLE tags that phones nearby report to the owner's account. Abusers hide them in a victim's bag or car. Apple and Google now alert people on iOS 17.5 and later and Android 6.0 and later when an unknown tracker is moving with them, under a shared specification being standardised at the IETF.

Trackers such as AirTag and Find Hub compatible tags are cheap, last a year on a coin cell, and use the millions of phones around them to report their location. That makes them excellent for finding keys and, in the wrong hands, for following a person.

What Apple and Google have done

In February 2022, Apple published an update on AirTag and unwanted tracking. It said incidents of misuse were rare but that it had seen bad actors attempt to misuse AirTag, and that it worked with law enforcement, since every AirTag has a serial number and paired AirTags are associated with an account. It announced plans for Precision Finding to locate an unknown AirTag, and for an on-screen alert when an AirTag that is moving with you plays its sound.

On 13 May 2024, Apple and Google shipped cross-platform alerts based on a joint industry specification called Detecting Unwanted Location Trackers. iOS 17.5 and Android 6.0 and later show an "[Item] Found Moving With You" alert when an unknown compatible tracker travels with you over time. Tag makers including Chipolo, eufy, Jio, Motorola and Pebblebee committed to compatible future tags.

Google's help page on finding unknown trackers adds two details worth knowing. The alerts currently work with Find Hub compatible tags, headphones and AirTags. And turning off Bluetooth, location or putting the phone in airplane mode does not stop someone else's tracker from sharing its location; you need to find it and follow the maker's disablement steps.

What the standard requires

The work continues in the IETF's Detecting Unwanted Location Trackers (DULT) working group. The current accessory protocol draft, version 01 from October 2026, sets rules for tracker makers:

  • A tracker near its owner rotates its address every 15 minutes, so strangers cannot follow the owner by scanning.
  • After it has been separated from its owner for more than 30 minutes, it moves to separated mode and rotates its address only every 24 hours. The longer window is deliberate: it lets a victim's phone notice that the same tag has been with them for a while.
  • It must include a sound maker that plays when separated and when a non-owner triggers it, at a minimum loudness of 60 phon.
  • A non-owner must be able to retrieve an identifier and get disablement instructions.

Notice the trade-off. The same rotation that protects owners from tracking would hide a stalking tag from its victim, so the standard rotates slowly only when the tag is away from its owner.

Detection and defence for people at risk

  • Keep your phone updated and leave unknown tracker alerts on.
  • On Android, you can also run a manual scan from the unknown tracker settings.
  • Researchers at the SEEMOO lab publish AirGuard, an open source app to detect AirTags and Find My accessories.
  • If you find a tracker, Google's guidance suggests going to a safe public location, and Apple's guidance points to support organisations for people whose safety is at risk. If you feel unsafe, contact local support services or the police before you disable it.

How secure are Bluetooth car keys and phone-as-a-key systems?

Bluetooth phone-as-a-key systems unlock a car when the owner's phone seems nearby, but a Bluetooth connection is not proof of distance. Link layer relay attacks can forward the connection with very little delay. Defences are secure ranging with UWB or Bluetooth Channel Sounding, user confirmation and motion-aware passive entry.

This section is conceptual by design. It explains why the risk exists and how it is reduced, not how to carry out an attack.

The relay problem

Passive entry works like this: the car sees the owner's phone, the phone proves it holds the right key, and because the connection exists the car infers the phone is close. A relay attack breaks the last step. One device near the owner and another near the car forward the radio traffic between them, so the car talks to the real phone and believes it is nearby.

In May 2022, NCC Group published a technical advisory on BLE proximity authentication. Its researchers built a relay that works at the link layer, adds as little as 8 milliseconds of round-trip delay (within the normal variation of BLE responses, given typical connection intervals of 30 milliseconds or more), and can forward encrypted link layer traffic. Two common defences, response time limits and link layer encryption, did not stop it. The advisory lists phone-as-a-key systems, residential smart locks, building access systems and laptops with trusted-device unlock as the kinds of products affected. NCC also pointed out that the Bluetooth Proximity Profile itself warns that proximity indicated by a BLE connection should not be the only protection of valuable assets.

What fixes it

NCC's recommendations line up with where the industry has gone:

  • Measure distance, not signal. Time-of-flight ranging cannot be faked by relaying, because a relay adds travel time. The Car Connectivity Consortium's Digital Key Release 3 combines Ultra Wideband (UWB) with BLE for hands-free, location-aware access, and the consortium runs a certification program for it. Bluetooth itself added Channel Sounding in Core 6.0, which measures distance with phase and round-trip timing; the SIG's technical overview includes sections on attack detection and round-trip time security.
  • Ask the user. Require a tap or biometric on the phone before unlocking or starting.
  • Use motion. NCC suggests disabling passive unlock when the phone has been stationary for more than a minute.
  • Offer an off switch. Let owners disable passive entry that relies on inferred proximity alone.

The head unit is a separate target

The access module is not the only Bluetooth surface in a car. The infotainment head unit pairs with phones to sync contacts and messages. PerfektBlue showed that a flaw in its Bluetooth stack can give code execution there, with access to location, microphone audio and contacts. Car owners should remove old phones from the car's paired list, avoid pairing with unknown devices, and accept vehicle software updates. Fleet operators and rental companies should clear paired phones between drivers.

Conceptual diagram of a relay attack on phone as a key, with normal use, the relay case and four defences

Figure 8. Phone as a key and the relay problem, shown conceptually, with the defences that measure distance or ask the user.

How do you detect a Bluetooth attack?

Bluetooth attacks rarely announce themselves, so detection relies on indirect signals: unexpected pairing prompts, unknown devices in the paired list, new keyboards, unknown tracker alerts, Bluetooth crashes or reboots, and in organisations, inventory and patch data that show which devices are exposed.

Signals on a single device

  • A pairing request or code confirmation you did not start. Decline it.
  • A device in your paired list you do not recognise, especially a keyboard or mouse.
  • Text appearing or actions happening on their own.
  • Earbuds or a watch that reboot, disconnect or hang repeatedly in public places, which is the visible symptom of crash and deadlock bugs like SweynTooth and BrakTooth.
  • An unknown tracker alert.

Signals for an organisation

  • Inventory and patch level. The strongest detection is knowing exposure before an attacker does. Which laptops run which Bluetooth chip and driver version? Which phones are below the patch level that fixed CVE-2023-45866? Which BLE locks and badge readers use which SoC?
  • Mobile device management (MDM). Managed phones and laptops can report Bluetooth state and paired devices, and enforce policies such as disallowing file transfer or limiting which device classes can pair.
  • Endpoint logs. Operating systems log new Bluetooth bonds and new HID devices. An alert on a newly paired keyboard on a sensitive workstation is cheap and useful.
  • Radio monitoring. In high-security areas, BLE and Classic sensors can inventory devices in range and flag unexpected ones. This is mostly useful for spotting unknown trackers and rogue devices, not for catching protocol attacks in progress.
  • Lab capture for product teams. Device makers can watch the host controller interface and over-the-air traffic of their own products in the lab, to verify key sizes, pairing methods and behaviour under malformed input. BlueZ's HCI monitor is an example of a tool used for this on Linux.

Why do Bluetooth flaws spread through the supply chain?

Most Bluetooth products do not write their own Bluetooth stack. They build on a chipset vendor's SDK or a third-party stack, often through a module supplier. One flaw in that shared code ships in thousands of products, and each fix must pass through every company in the chain before it reaches users.

This is the part of Bluetooth security that buyers see least and that causes the longest exposure.

The chain

  1. A chipset vendor ships an SoC with an SDK: controller firmware, often a BLE host stack, sample applications and update tooling.
  2. A module maker or design house wraps the chip in a certified module or a reference design and sometimes adds its own firmware.
  3. A device maker builds the earbuds, watch, lock or head unit on that module, adds brand features and a companion app.
  4. An update channel, a phone app, an OTA server or a dealer visit, carries any fix to the device.
  5. The user installs it, if they ever open the app.

Flow diagram of how one chipset SDK flaw spreads to many products and how the fix must travel back, with SweynTooth, BrakTooth, Airoha and PerfektBlue as examples

Figure 9. One SDK, thousands of products. The flaw ships wherever the SDK went; the fix travels back one company at a time.

The public record

The cases in this guide show the pattern four times.

  • SweynTooth. Twelve flaws in BLE SDKs from seven SoC vendors, with about 480 product listings using the affected chips. The SoC vendors patched during disclosure; the products needed their own updates.
  • BrakTooth. Sixteen flaws in Classic chip firmware, more than 1,400 product listings, and a warning that because stacks are shared, many more products were probably affected.
  • Airoha. One SDK's unauthenticated debug protocol inside earbuds and headphones from many brands, some of which did not know whose chip they were using.
  • PerfektBlue. One third-party stack in cars from several makers, with a patch released in September 2024 that some carmakers still had not received by June 2025.

Why it is hard

Several things make Bluetooth supply chains harder than ordinary software:

  • Closed source firmware. Controller firmware is binary. Device makers often cannot inspect it, and buyers never can.
  • Opaque bills of materials. A device maker may know the module part number but not the chip, the SDK version or the third-party stack inside it.
  • Qualification is not security testing. The SweynTooth researchers called out concrete flaws in the BLE certification process. Qualification checks conformance to the specification, not resistance to malformed input.
  • Slow or absent update channels. Many cheap BLE devices have no update mechanism at all. Many that do rely on a companion app the user rarely opens.
  • Long product lives. Cars, medical devices and building access systems stay in service for many years.

A public lookup that helps

The Bluetooth SIG's qualified product listings tie products to the qualification IDs of the chips and stacks they use. Both SweynTooth and BrakTooth researchers used those listings to estimate how many products were affected, and PCA suggests the same approach for working out whether a device uses BlueSDK. A device maker can use the same records to check their own exposure, and a buyer can use them to ask sharper questions.

Regulation is catching up

In the EU, the Cyber Resilience Act entered into force on 10 December 2024. Its reporting obligations apply from 11 September 2026 and its main obligations from 11 December 2027. It requires manufacturers of products with digital elements to handle vulnerabilities across the product lifecycle and to meet cybersecurity requirements at every stage of the value chain. For a Bluetooth product, that means knowing what is inside the firmware, patching it, and being able to prove both. Our post on the EU Cyber Resilience Act and SBOM obligations covers the details.

How should home users secure Bluetooth?

Keep phones, laptops and accessories updated, turn Bluetooth off when you do not need it, only pair when you started the pairing, compare codes before accepting, remove old pairings, and act on unknown tracker alerts. These steps close most of the practical attacks in this guide.

  1. Update the operating system on your phone and laptop promptly. BlueBorne, BlueFrag, BleedingTooth and CVE-2023-45866 were all fixed through OS updates.
  2. Update your accessories. Earbuds, headphones, watches and trackers get firmware through their companion apps. Open the app now and then and install updates. This is how fixes for chip flaws such as the Airoha issues reach you.
  3. Turn Bluetooth off when you are not using it, especially on laptops and in crowded places. The less time a device spends listening, the less time an attacker has.
  4. Only pair when you started it. Decline any pairing request or code confirmation that pops up unprompted.
  5. Compare the six digit code when your phone and another device both show one. That is what gives Numeric Comparison its protection.
  6. Pair in private. Pair new earbuds at home rather than on a train. Some attacks need to be present at pairing time.
  7. Remove pairings you no longer use, including old cars, rental cars and speakers.
  8. Check for strangers in your paired list, especially keyboards and mice you do not own.
  9. Keep unknown tracker alerts on, and if you get one, follow the platform's steps to find the tracker. Think about your safety before you disable it.
  10. Avoid personal names in device names. "Priya's AirPods" broadcasts more than it needs to.
  11. For cars with phone-as-a-key, install vehicle updates, use any option to require phone confirmation, and turn off passive entry if you do not need it.
  12. Retire devices that no longer get updates. If a maker has stopped shipping firmware for a Bluetooth device you use for anything sensitive, replace it. Our guide on how long a smart device gets security updates shows how to check.

How should businesses secure Bluetooth?

Businesses should inventory every Bluetooth device and chipset they own, enforce Bluetooth policy through device management, alert on new input devices, track controller and accessory firmware alongside OS patches, review Bluetooth locks and phone-as-a-key systems for relay risk, and demand firmware SBOMs from suppliers.

  1. Build an inventory. List laptops, phones, headsets, conference room systems, badge readers, smart locks, medical and lab equipment, and vehicles, with their Bluetooth chips, drivers and firmware versions. You cannot patch what you have not listed.
  2. Set Bluetooth policy through MDM. Decide per device class whether Bluetooth is allowed, which profiles are permitted, whether file transfer is allowed, and whether discoverable mode is ever acceptable.
  3. Alert on new input devices. New Bluetooth keyboards and mice on sensitive machines should raise an alert. CVE-2023-45866 shows why.
  4. Patch more than the OS. Track Bluetooth driver and controller firmware updates for laptops, and firmware for headsets and conference equipment. They are often outside the normal patch pipeline.
  5. Prefer devices that support Secure Connections Only mode, and turn it on where the platform allows.
  6. Review physical access systems. For BLE locks, badge systems and fleet vehicles that use phone-as-a-key, ask vendors how they resist link layer relay attacks. Distance bounding with UWB or Channel Sounding, user confirmation and motion checks are the answers to look for.
  7. Handle trackers. Brief staff, especially those who travel or who face threats, on unknown tracker alerts. Sweep company vehicles periodically.
  8. Treat BLE beacons and analytics as data processing. If you log visitor devices, apply your privacy rules.
  9. Buy with security in the contract. Ask suppliers for firmware SBOMs, the chipset and stack versions inside, the update policy and support period, and their disclosure process. Our post on third-party risk management with Safeguard TPRM covers how to run this as a program.
  10. Train people. Most practical Bluetooth attacks need a user to accept something, or to not notice something. Teach staff to decline unsolicited pairing requests and report odd prompts.

Defence in depth diagram with six Bluetooth layers: exposure, pairing, patching, services, monitoring and supply chain, and the attacks each helps against

Figure 10. Defence in depth for Bluetooth. Each layer covers gaps in the others.

How should device makers secure Bluetooth?

Device makers should require authenticated pairing for anything sensitive, support Secure Connections Only mode, enforce minimum key sizes, remove debug protocols from production, validate every key exchange input, know exactly which chip SDK and stack versions they ship, fuzz the stack, ship signed updates for the product's life and run a real disclosure process.

This is the longest checklist because device makers control the most.

Pairing and cryptography

  1. Never use Just Works for control features. If a GATT write can unlock a door, change insulin settings or start a car, it must require authenticated pairing.
  2. Support Secure Connections Only mode and enable it wherever the use case allows. It blocks the downgrades behind BIAS and BLUFFS.
  3. Enforce minimum key sizes: reject Classic links below 7 octets and, where you always use Security Mode 4 Level 4, below 16 octets.
  4. Remember how a bond was made. If a peer bonded with Secure Connections, terminate any later connection that tries to use legacy encryption, as the SIG recommends after BLUFFS.
  5. Restrict cross-transport key derivation as specified in Core 5.1 and later, and never let a weaker key overwrite a stronger one.
  6. Validate public keys in every ECDH exchange.
  7. Limit pairing windows. Enter pairing mode only on a deliberate user action and only for a short time.

Services and protocols

  1. Authenticate every GATT characteristic that changes state or exposes personal data.
  2. Remove, disable or authenticate debug and tuning protocols before production. Airoha's flaws were a debug-capable protocol open to anyone in range.
  3. Require explicit user consent before accepting HID connections in host stacks, and default security fixes to on.
  4. Treat the controller as untrusted from the host's side. BleedingTooth showed a malicious chip can trigger host bugs.

Privacy

  1. Rotate addresses and all identifying advertisement fields together.
  2. Stop advertising when not needed and use Encrypted Advertising Data for sensitive payloads.
  3. For trackers, implement the DULT accessory protocol, including separated mode, the sound maker and identifier retrieval.

Proximity and access

  1. Do not rely on connection presence or signal strength as proof of distance. Use UWB or Bluetooth Channel Sounding with its security features, or require user confirmation.

Supply chain, testing and updates

  1. Know what you ship. Record the SoC, controller firmware version, SDK version, host stack and every open source component in a software bill of materials (SBOM) for each firmware release.
  2. Subscribe to your suppliers' security advisories and to the research groups that publish Bluetooth disclosures.
  3. Fuzz the stack and test against public proof-of-concept cases such as SweynTooth, BrakTooth and the BLUFFS checker in your own lab.
  4. Ship signed OTA updates and keep shipping them for the stated support period.
  5. Publish a vulnerability disclosure policy and patch timelines, and tell customers when a shared SDK flaw affects them.

Three column checklist graphic for Bluetooth hardening for home users, businesses and device makers

Figure 11. The hardening checklist in one view.

How Safeguard helps teams that build Bluetooth products

Safeguard covers the software supply chain inside a Bluetooth product in one platform: SBOMs for firmware and companion apps, binary and image scanning, vulnerabilities ranked by reachability, SCA, SAST and secrets scanning on code you own, fix pull requests gated by your CI, Gold patched packages, supplier risk and compliance evidence.

Safeguard does not sit in the radio. It sits in the place where most long-lived Bluetooth exposure comes from: the firmware, SDKs, open source components, companion apps and update services that a product team ships and maintains. Teams that take this seriously often end up with separate tools for source scanning, binary analysis, SBOM management, supplier questionnaires and audit evidence. Safeguard does that work end to end with one inventory and one vulnerability model, from the first SBOM to the merged fix and the audit record.

Flow diagram of Safeguard for Bluetooth products: source repos, app binaries, images, extracted firmware and supplier SBOMs feed SBOMs, reachability-ranked vulnerabilities and code scanning, leading to fix pull requests, Gold packages, supplier risk, compliance evidence and SBOM delivery

Figure 12. How Safeguard fits a Bluetooth product, from inputs to fixes and evidence.

See inside the binaries and images you ship and receive

In the Safeguard console, Integrations > Upload App Binary accepts compiled software: Windows EXE, MSI and DLL files, Android APKs and iOS IPAs, VM and disk images in QCOW2, VMDK, VHD and OVA formats, and container tarballs. Safeguard unpacks the file, identifies embedded libraries and their versions, maps them to known vulnerabilities and records licences, and the result becomes a project with a full SBOM. For a Bluetooth product, that covers the companion app on both mobile platforms, desktop pairing and update utilities, and Linux-based images for head units, gateways and hubs where the BlueZ and kernel versions matter. Raw firmware images for microcontrollers and SoCs may need to be extracted into their component files first. The usual limits of binary analysis apply: heavy obfuscation, encryption or unusual compression can give incomplete results. More detail is in our guide to scanning app binaries with Safeguard and our explainer on binary composition analysis.

Rank vulnerabilities by what can actually be reached

A firmware image or companion app can pull in hundreds of components, and many vulnerabilities in them are not reachable from your code. Safeguard's reachability analysis ranks vulnerabilities your code can actually reach above those it cannot, and adds EPSS (the Exploit Prediction Scoring System) and known exploited context, so a team can tell the BlueZ flaw on an internet-connected hub apart from an unused library function.

Scan the code you own

For firmware, companion apps and cloud services you build yourself, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree, static application security testing (SAST) on your own code, and secrets scanning that catches signing keys and API tokens before they ship. For a device maker, a leaked OTA signing key turns a software bug into a fleet-wide compromise, so secrets in firmware repositories are worth catching early.

SBOMs your customers and regulators can use

Every project produces an SBOM you can export in CycloneDX or SPDX, among other formats. That is what a business buyer, a carmaker or an auditor will ask for, and what lets you answer "do we ship BlueSDK, which Airoha SDK, which BlueZ version" in minutes. Safeguard's Third-Party Risk Manager lets you request SBOMs from your own chip and module suppliers, score them and monitor them continuously, so the supplier side of the chain is visible too.

Fix it without waiting for upstream

This is where Safeguard differs most from a scanner. For code in your repositories, Griffin, Safeguard's remediation model, opens a fix pull request that your own CI builds and tests before anyone merges it, and your policies decide what can merge. For open source components where upstream has not shipped a fix, or the fix sits in a version you cannot move to without a larger change, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next build pulls a fixed version. For a device team waiting on a supplier's release schedule for the open source parts of its firmware or app, that is the difference between shipping the fix in the next build and waiting a quarter. The closed source controller firmware still has to come from the chip vendor, but everything around it no longer has to wait.

Compliance evidence from the same data

Safeguard's compliance suite collects evidence for frameworks such as SOC 2 and ISO 27001 from the same scans, policies and fix history. For teams preparing for obligations like the EU Cyber Resilience Act, the SBOMs, vulnerability handling records and fix trail are the raw material that audits and regulators ask for.

Deploy where your code lives

Firmware source and signing infrastructure are sensitive. Safeguard runs as SaaS, in your own VPC, or fully on premises. It can run in Zero AI mode with no language model in the loop, or with a model you choose.

How this compares

Device teams often pair a source SCA tool such as Snyk or Black Duck with a firmware and product security specialist such as Finite State, which describes itself as securing every device, and a GRC platform for audits. Those are capable products, and for deep analysis of raw firmware images a firmware specialist will often go further. Safeguard's case is consolidation and remediation: one inventory across source, binaries and images; one vulnerability model with reachability; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. Safeguard can also work alongside an existing firmware analysis tool and take over SBOM delivery and fixing. For more context, see our firmware supply chain security guide, SBOMs for firmware and IoT devices and software supply chain security for automotive.

The takeaway

Bluetooth security is mostly not about exotic radio tricks. The protocol flaws are real and clever, but the attacks that reach people are simpler: a debug protocol open to anyone in range, a stack that pairs a keyboard without asking, a host parser with a memory bug, a car that trusts a connection as proof of distance, and a tracker in a coat pocket.

The fixes exist for nearly all of them. The hard part is getting the fix from the company that wrote the code to the device in someone's ear, on their wrist or in their driveway. That journey runs through chip vendors, module makers, device makers, apps and dealers, and it can take months or never finish. Users can shorten it by updating. Businesses can shorten it by knowing what they own. Device makers can shorten it most of all, by knowing exactly what is inside their firmware and being able to ship a fix without waiting for everyone upstream.

Try it yourself

Click through these interactive Safeguard demos to see the steps described in this guide.

Test yourself

What is the difference between Bluetooth Classic and Bluetooth Low Energy?

Classic (BR/EDR) carries audio, calls and keyboards. Low Energy (BLE) carries small bursts of data for wearables, trackers, locks and sensors, and announces devices with short broadcast advertisements.

Which pairing modes are weak, and why is Just Works a problem?

Legacy PIN pairing, LE legacy pairing and Just Works are the weak ones. Just Works gives no protection against a man in the middle, and devices with no screen, like earbuds, fall back to it, which the post calls the most common weakness in consumer Bluetooth.

What did the KNOB attack abuse?

Bluetooth Classic lets two devices negotiate an encryption key length from 1 to 16 bytes, and that negotiation was not integrity protected. An attacker could lower it to one byte, leaving only 256 possible keys to brute force.

What do most attackers need to hack Bluetooth?

Physical proximity, inexpensive hardware, and either a known flaw on an unpatched device or a weak configuration. Almost none of the attacks can be launched from across the internet.

What happened with CVE-2023-45866?

Several Bluetooth stacks could be tricked into pairing with a fake keyboard without user confirmation and then accept its keystrokes. The fixes came through operating system updates, and a good policy is that input devices may only pair after explicit user action.

How do Apple and Google help people spot unknown trackers?

Since 13 May 2024, iOS 17.5 and Android 6.0 and later show a Found Moving With You alert when an unknown compatible tracker travels with you. Turning off Bluetooth or location does not stop the tracker from sharing its location, so you need to find it.

Why do Bluetooth flaws spread through the supply chain?

Most products build on a chipset vendor's SDK or a third-party stack, often through a module supplier. One flaw in that shared code ships in thousands of products, and each fix must pass through every company in the chain before it reaches users.

Frequently asked questions

Can someone hack my phone through Bluetooth?

Yes, if the phone is missing security updates and the attacker is within radio range. BlueBorne in 2017, BlueFrag in 2020 and CVE-2023-45866 in 2023 all allowed attacks on phones without the user's help. Each was fixed by an operating system update, so keeping your phone updated is the most important defence.

Is it safe to leave Bluetooth on all the time?

On an up-to-date phone, leaving Bluetooth on is a reasonable trade-off for most people, and features like unknown tracker alerts need it. Turning it off when you do not need it, especially on laptops and in crowded places, still reduces exposure. Never accept pairing requests you did not start.

What is the most secure Bluetooth pairing method?

Numeric Comparison or Passkey Entry over Secure Connections, with both devices in Secure Connections Only mode. Numeric Comparison only protects you if you actually compare the six digit numbers. Just Works and legacy pairing give no protection against a man in the middle.

Can Bluetooth headphones and earbuds be hacked?

Yes. In 2025, ERNW showed that headphones and earbuds built on Airoha chips exposed a protocol that let anyone in range read and write device memory without pairing, and listen through the microphone. BrakTooth in 2021 also crashed several audio chips. Keep the headphone app updated so firmware fixes reach your device.

How do I know if an AirTag or other tracker is following me?

On iOS 17.5 or later and Android 6.0 or later, your phone alerts you when an unknown compatible tracker is moving with you. You can then play a sound to find it and view disablement instructions. On Android you can also scan manually. If you feel unsafe, contact local support services before disabling the tracker.

Can thieves unlock a car that uses my phone as a key?

Systems that unlock just because your phone is connected nearby can be vulnerable to relay attacks, which forward the Bluetooth connection between your phone and the car. Cars that measure distance with Ultra Wideband or Bluetooth Channel Sounding, or that require a tap on the phone, are much harder to fool. Install vehicle updates and use those options where offered.

Does BLE address randomization stop tracking?

It helps but does not solve it alone. BLE privacy changes the device address regularly so passive scanners cannot link sightings by address. Research from Boston University showed that stable fields in the advertising payload can still let a scanner follow a device across address changes, so makers must rotate everything together.

How should device makers secure Bluetooth firmware?

Require authenticated pairing for sensitive functions, support Secure Connections Only mode, enforce minimum key sizes, remove debug protocols from production, and validate every key exchange input. Above all, record the chip SDK, stack and component versions in an SBOM for each release, scan your firmware and apps, and ship signed updates for the product's whole life.

Sources

All sources are listed in the front matter of this post with exact URLs. Key primary sources:

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.