Blog
Security insights.
Research, analysis, and operational lessons from the front lines of AI agent security.
Amazon’s AI Outage Wasn’t the Problem
Amazon added senior engineer sign-offs after its AI outage. That’s not security. It’s safety performance. The real failure was configuration — and configuration is where the fix has to live.
ChatGPT Practiced Law for Months. Nobody Noticed.
A federal court in Chicago is now the place where the AI configuration argument gets made in public. This isn't about a hallucination. It's about drift — and the security model that couldn't see it coming.
Your AI Agent Is Learning Everything About Your Business. So Is Everyone Else's.
Managed AI agent platforms promise to handle your operations. What they don't tell you: every decision those agents make trains models that serve your competitors on the same platform. Before you connect your systems, understand what you're really signing up for.
Sandboxing Is Not Security: What Perplexity Computer Gets Wrong
Perplexity launched a 19-model AI agent orchestrator with persistent memory and hundreds of integrations. Their security answer is a sandbox. That solves containment. It doesn't solve prevention.
Thousands Lost Their Google Accounts This Month. Their AI Agents Were the Reason.
A popular AI agent framework crossed a provider's Terms of Service. Users didn't just lose their AI tools — they lost Gmail, Workspace, Drive, and YouTube. The configuration layer is the security surface nobody's testing.