Safeguard
Topic

AI Security

In-depth guides and analysis on ai security from the Safeguard engineering team.

676 articles

AI Security

Training Data Opacity As A Trust Limit

You cannot audit what you cannot see. Frontier model training corpora are effectively opaque to their users, and that opacity is not incidental. It shapes what kinds of trust you can extend to the outputs.

Jan 31, 20267 min read
AI Security

Griffin AI vs Gemini Long Context for Codebases

Gemini's million-token context window is a genuinely new capability. For security analysis of large codebases, is it enough on its own?

Jan 31, 20267 min read
AI Security

Human Review Burden: Griffin AI vs Mythos

Auto-remediation only scales if human review stays cheap. Griffin AI's grounded PRs keep reviewer time low; Mythos-class PRs push the cost back to humans.

Jan 31, 20266 min read
AI Security

Agent Security: Enterprise Adoption Patterns

Enterprise agent deployments have moved past pilot phase. The security patterns that have survived contact with production look different from the ones the industry was selling a year ago.

Jan 30, 20267 min read
AI Security

Claude Code Coding Agent: Security Posture Review

A working review of Claude Code's security posture, sandboxing model, and the practical controls enterprises need to deploy it safely at scale.

Jan 30, 20265 min read
AI Security

Cross-Package Analysis: Griffin AI vs Mythos

Real exploits cross package boundaries. Griffin AI's graph follows them; Mythos-class tools often stop at the file they are reading.

Jan 30, 20266 min read
AI Security

AI-BOM Awareness: Griffin AI vs Mythos

AI-BOM is how you describe an AI system's supply chain — models, datasets, prompts, inference environments. Griffin AI ingests it as structured inventory. Mythos-class tools try to talk about AI while remaining blind to the AI systems they describe.

Jan 29, 20267 min read
AI Security

SOC 2 Type II Evidence: Griffin AI vs Mythos

A SOC 2 Type II auditor samples a control population across a reporting period. Griffin AI creates that population as a natural output. Mythos-class pure-LLM tools leave you reconstructing it.

Jan 29, 20267 min read
AI Security

AI Model Weight Tampering Detection Techniques

Weight-level tampering leaves cryptographic and statistical fingerprints. Here is what current research says about detecting a modified checkpoint before it reaches inference.

Jan 29, 20268 min read
AI Security

Sandboxing LLM Agent Code Execution: Patterns

If your agent can execute code, something it reads from the internet can execute code. Pick your sandbox before the agent picks one for you.

Jan 29, 20267 min read
AI Security

Citation Accuracy: Griffin AI vs Mythos

An AI security tool that cites the wrong advisory is worse than one that says nothing. Griffin AI benchmarks citation accuracy at 0.89 similarity; Mythos does not.

Jan 28, 20267 min read
AI Security

SSRF Detection: Griffin AI vs Mythos

Server-side request forgery is a test of how well your scanner understands the boundary between trusted and untrusted URLs. Griffin's engine resolves URL construction through string builders, template engines, and HTTP client configuration; Mythos reads the code and guesses. On modern applications that is the difference between a finding you can ship and a finding you cannot defend.

Jan 28, 20267 min read
AI Security (Page 38) — Supply Chain Security Blog | Safeguard