Flaws and attacks in the tools AI runs on: model servers, ML frameworks, agents and MCP servers, model hubs, prompt injection. Where the AI risk index and the security watch meet.