Detecting the two threats nobody else is covering:
AI-accelerated insiders and autonomous agents gone wild.
Who watches the agents?
Malicious actors using Claude Code, Cursor, or automation platforms to exfiltrate entire codebases in a single session. The intent is human. The velocity is machine.
Autonomous agents that drift beyond intent -- deleting emails, exposing API keys, following prompt-injected instructions from malicious documents. The agent becomes an insider threat to the very person who launched it.
Traditional insider threat detection was built for human-speed attacks. That era is over. FrawdBot is a 21,000-line Python detection engine running 12 behavioral rules against rolling statistical baselines, with campaign detection that connects the dots across weeks of activity. Launching on Google Workspace, with Microsoft 365 integration in development.
Part of the OpenClaw ecosystem. Built by Self-Improving Code.
selfimprovingcode.ai →