RESEARCH
Attack chains, not generic commentary
We focus on concrete exploit paths: prompt injection, tool abuse, permission overreach, leaked secrets, and exposed endpoints.
Read the briefings →AI SECURITY HUB
Learn. Break. Build. Defend.
Hands-on AI security across research, public exposure tracking, and release infrastructure for real Agent systems.
Prompt InjectionAgent AbuseRAG PoisoningTool CallingOpenClaw Exposure
Researchattack research and defensive breakdowns
Watchboardspublic evidence surfaces for exposed Agent targets
Infrapolicy, scan, and release gate control plane
SIGNALS
How AIPwn works
The site should not read like a thin company page. It should show a loop: find, explain, and operationalize.
WATCHBOARDS
Public proof that scanning is happening
OpenClaw watchboards make security results visible, sortable, and shareable instead of burying them in a single report artifact.
Open OpenClaw →PRODUCT
Turn findings into release control
ClawPlane connects policy, scan, and gate decisions so the same evidence flows into CI, deploy, and public trust surfaces.
Open ClawPlane →PRODUCTS
What we are building
Everything orbits one goal: make AI systems harder to break and easier to verify.
Newsletter
AI security briefings with real attack chains, defense breakdowns, and practical industry signals.
LiveClawPlane
Agent security and release infrastructure. One control plane for policy, scanning, and CI/deploy gates.
AlphaClawScan
Evidence-rich scanning for repos, skills, MCP servers, and exposed OpenClaw services.
Live SurfaceClawGate
PR, CI, and deploy gate module that blocks risky merges based on policy thresholds and diff-first findings.
AlphaCOVERAGE
What we track across the Agent attack surface
The homepage should already tell visitors what problems AIPwn actually understands.
Current focus areas
Prompt Injection
Instruction override, hidden tool abuse, indirect prompt poisoning, and unsafe retrieval flows.
Tool And Runtime Abuse
Shell execution, downloader chains, unsafe subprocess usage, and overly broad permissions.
Secret Exposure
Leaked API keys, tokens in repos, unsafe logs, and public config artifacts.
Public Agent Exposure
OpenClaw services exposed without auth, public docs/openapi, and externally reachable risky interfaces.