-
The Agent has Credentials: Agentic AI as Attack Surface in Financial Services
17,600 attacker actions in 4.5 days, no operator. What the Hugging Face timeline and GTG-1002 mean for banks, and what to test now.
-
Your AI agent is reliably breakable. We tested it.
June 2026 research confirms frontier LLMs break under automated pressure. For agents with tool access, that is no longer a PR problem.
-
The MCP server attack surface
MCP servers are running on engineering laptops at most companies, holding production credentials, with no inventory. Here’s the threat model and what to do.