AI Security
Sharing a Claude chat link: what people see and how to unshare
A shared Claude chat is a snapshot that anyone with the link can read. What it includes, what stays private, and how to revoke a link on any plan.
Oct 18, 20263 min read
Resources
Supply Chain Security, in plain English.
Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
Filter
Articles
RSS feedAI Security
Claude voice dictation: is your audio kept or used for training?
Anthropic says Claude mobile dictation audio is deleted after speech becomes text, and your voice is not used for training. The transcript is different.
Oct 18, 20263 min read
AI Security
How long Claude keeps your chats: 30 days, 2 years and 5 years
Anthropic's consumer retention rules in plain words: deleted chats, the training setting, safety flags and feedback, and which clock applies to you.
Oct 18, 20263 min read
AI Security
How do you check the CVE and CWE references an AI assistant gives?
Models can invent identifiers that look real. Check each CVE and CWE an assistant cites against the primary record, and compare it with the claim.
Oct 18, 20264 min read
AI Security
Changing the model behind a production agent: treat it as a release
A new model version changes agent behavior even if your code stays the same. A change control routine: pin the version, test, canary and roll back fast.
Oct 18, 20263 min read
AI Security
Can I ask Claude or ChatGPT about my vulnerabilities?
Yes, if the assistant is connected to your scan data through MCP. Without a connection it can only guess. Here is how it works and what to lock down.
Oct 18, 20263 min read
AI Security
Can an MCP server ask you for a password or API key?
No. The MCP spec bans form mode requests for passwords and API keys and requires URL mode instead. Here is what each mode does and what clients must show.
Oct 18, 20264 min read
AI Security
Can an AI coding assistant use a security scanner through MCP?
How MCP lets an AI assistant call a security scanner, what to check before connecting one, and how to keep it from changing more than you intend.
Oct 18, 20263 min read
AI Security
Can a local app pretend to be your MCP client on localhost?
Yes. A local process can claim a known client identity and catch the sign in redirect on a localhost port. Here is how the attack works and what to check.
Oct 18, 20263 min read
Stay informed
Weekly insights on software supply chain security, delivered to your inbox.
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.