llm
Safeguard articles tagged "llm" — guides, analysis, and best practices for software supply chain and application security.
40 articles
LLM-assisted vulnerability autofixing: approaches and how to validate the patches
At DARPA's AIxCC finals in August 2025, AI systems patched 68% of vulnerabilities they found — up from 25% at semifinals. Here's how the approaches differ and why validation still matters most.
AI Code Review Tools Compared: An Honest 2026 Guide
A balanced 2026 comparison of AI code review tools — GitHub Copilot, CodeRabbit, Qodo, Graphite, Amazon Q, Snyk DeepCode — with honest tradeoffs, the security gap, and where Safeguard fits.
Claude Code Skills: A Security Guide to SKILL.md and Agent Extensions
Claude Code skills package instructions and scripts an AI agent runs on your behalf. That power is also the risk. Here is how to vet and sandbox them.
Building an Eval Suite for Your Security LLM Workflows
If you use an LLM anywhere in your security program — triage, remediation, detection — you need an eval suite with the same rigor as your test suite. Here is a concrete harness: datasets, thresholds, CI gates, and drift detection.
Zero-Day Discovery With LLM-Augmented Reachability: A Safeguard Engine Walkthrough
Pattern-matching scanners miss zero-days by definition. An engine that follows taint across package boundaries plus a model that hypothesizes exploit conditions can find what either would miss alone. Here is how that pipeline works end to end.
Frontier LLM Vendors Are Not Your Supply Chain Security Vendor
Coding agents from OpenAI, Anthropic, and Google are excellent tools. They are also not supply chain security platforms, and the assumption that they can replace one is already producing expensive gaps.
MCP Meaning: What the Model Context Protocol Is and Why It Matters
The MCP meaning most people are asking about is the Model Context Protocol, an open standard that lets AI models connect to tools and data through one common interface.
Why LLMs Are Structurally Insecure (and What That Means for Your Pipeline)
Language models are not insecure because of a bug you can patch. They are insecure by construction — non-deterministic, context-poisonable, and unreproducible. Here is how to reason about them without pretending otherwise.
The Limits of Single-Model Vulnerability Scanning: A Technical Analysis of the Mythos Approach
Anthropic's Mythos model claims to find vulnerabilities in open-source code using a single LLM. We analyze where this approach falls short and why production-grade zero-day discovery requires Safeguard's Multi-Agent TAOR Deep Think AI Engine.
Why LLM-Based Vulnerability Scanning Needs More Than a Single Model
Large language models are being used to find vulnerabilities in open-source code. But a single model, no matter how capable, isn't enough. Here's why multi-agent orchestration, structured CWE analysis, and deep context matter more than model size.
LLM Traces and Evals: The Missing Layer in AI Supply Chain Security
Prompt traces and offline evals are standard hygiene for ML teams, but almost nobody treats them as supply chain telemetry. They should be. Here's how traces and evals plug into SBOM and reachability as a fourth security signal.
Prompt Injection in RAG: Indirect Attacks
A senior engineer's breakdown of indirect prompt injection in RAG pipelines, how real attacks land through retrieved content, and what actually reduces exposure.
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.