Security Feed
Signals for agent runtime security.
Current AgentKeeper research, advisories, and operational notes for teams securing AI coding and productivity agents.
Runtime enforcement
Risk is evaluated when an agent tries to act, not only when a prompt is submitted.
MCP control plane
MCP server, tool, arguments, identity, and verdict belong in the same policy path.
Investigation chain
Prompt, tool input, output, model, repository, workstation, and identity stay correlated.
Latest notes
Field research
Field note | 4 minRuntime decisions beat prompt reviewPrompt review is useful context. The durable control point is the moment an agent tries to read, write, run, fetch, or call a tool.ReadArchitecture note | 5 minMCP is the productivity-agent control pointProductivity agents do real work through MCP tools. Those calls need the same policy path as coding-agent tools.ReadInvestigation note | 4 minInvestigations need a complete chainA useful security timeline connects prompt, tool input, output, model, cost, repository, workstation, identity, and verdict.Read