AI Security
In-depth guides and analysis on ai security from the Safeguard engineering team.
676 articles
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.
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?
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.