methodology · 3 min read
Every provider, one test surface
OpenAI, Anthropic, Google, Azure, Mistral, Groq, xAI, DeepSeek, Cohere, Bedrock, self-hosted. One adapter, consistent findings.
Resources
Written for the people who have to decide, not only the people who build. No payloads, no jargon for its own sake.
methodology · 3 min read
OpenAI, Anthropic, Google, Azure, Mistral, Groq, xAI, DeepSeek, Cohere, Bedrock, self-hosted. One adapter, consistent findings.
methodology · 4 min read
Every finding is one case: what was tried, the pattern at altitude, the outcome, what leaked, and the fix.
defences · 4 min read
The single most effective control for AI infrastructure is also the oldest: give each key the least it needs.
attacks · 5 min read
Copilot, Gemini and support agents read inboxes. A message can carry instructions addressed to them, invisible to the person.
attacks · 5 min read
An AI-backed endpoint carries context: the system prompt, the records it looked up, sometimes upstream keys. The right request can make it hand that back.
fundamentals · 6 min read
A model API key is treated like a password, but it behaves like a filesystem credential. Here is what someone holding it can reach before they ever send a prompt.