RESEARCH
Real exploit paths, not AI security hot takes
We break down prompt injection, tool abuse, secret leaks, and exposed agent endpoints into concrete attack chains teams can actually fix.
Subscribe Free →AI SECURITY RESEARCH
Learn. Break. Build. Defend.
Track AI security shifts, expose agent risks, and stop risky releases.
Prompt InjectionAgent AbuseRAG PoisoningTool CallingOpenClaw Exposure
Researchexploit paths and defensive breakdowns teams can act on
Evidencepublic watchboards for exposed agent targets
Controlpolicy, scanning, and release gates in one plane
HOW IT WORKS
From exploit path to release decision
AIPwn connects research that explains AI and agent attack paths, watchboards that prove exposure, and release controls that stop risky changes from shipping.
WATCHBOARDS
Public evidence that exposure is real
OpenClaw turns findings into visible watchboards so teams can verify what is exposed, where it is reachable, and how risk changes over time.
Open OpenClaw →PRODUCT
Release gates tied to real findings
ClawPlane uses the same evidence in policy, CI, and deploy decisions so risky changes do not quietly ship.
Open ClawPlane →PRODUCTS
What we are building
One stack for teams that need research, scanning, and release control across AI systems and agents.
AIPwn Newsletter
AI security briefings with real attack chains, agent exploits, defensive breakdowns, and high-signal industry shifts.
LiveClawPlane
Policy, scanning, and release gates for AI systems and agent workflows in one control plane.
AlphaClawScan
Evidence-first scanning for repos, skills, MCP servers, and exposed OpenClaw targets.
Live SurfaceClawGate
Diff-aware PR, CI, and deploy gates that block risky changes with evidence instead of guesswork.
AlphaCOVERAGE
What we detect across AI systems and agent surfaces
The homepage should make our detection surface obvious: prompt exploits, model and tool abuse, leaked secrets, and public exposure.
Current detection areas
Prompt Injection
Instruction override, hidden tool abuse, indirect prompt poisoning, and unsafe retrieval flows.
Tool Abuse And Runtime Risk
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 Endpoint Exposure
OpenClaw services exposed without auth, public docs/openapi, and externally reachable risky interfaces.