Field notes on shipping agents you can trust

Inside the technical report: how adaptive, coverage-guided testing found an order of magnitude more issues than expert manual testing on a 1.5M-deployment agent.

A breakdown of our AgentDojo results across Workspace, Slack, Banking and Travel — and why hard-tier tasks fall hardest when you see the loop.

A class of violation where an agent exfiltrates data only after touching a sensitive record. Why checklists miss it and how adaptive testing catches it.

Employees are wiring agents to ungoverned MCP servers. Why the shadow channel is the biggest blind spot in agent security today.

A court held an airline liable for its chatbot’s invented refund rule. What that precedent means for every enterprise deploying agents.

Verigrey has been accepted into Cyberboost, the Cyber Security Agency’s highly selective accelerator program for cybersecurity startups.

A practical guide to turning fuzzy guardrails into precise, checkable sentences your whole team agrees on.

The evidence to demand before an autonomous AI agent reaches production — seven domains mapped to OWASP, MITRE ATLAS, NIST AI RMF, and the regulations auditors already use.

A running log of platform releases, benchmark updates, and new attack-class coverage is on its way. Check back soon for updates the moment they ship.

New LLM, new prompt, new tool call — any of them can silently break assurance. How to trigger a new pass automatically in CI/CD.