01
Latest briefPROMPT INJECTION
Prompt injection is now a tooling problem
The break lands through retrieval, orchestration, and hidden instructions before it ever looks like a jailbreak.
Subscribe FreeAI SECURITY BRIEF
AI security for teams in production.
AIPwn tracks how AI systems break: exploit paths, exposed surfaces, and ship risk.
Weekly research for teams that need signal.
Prompt InjectionTool AbusePublic ExposureRelease Control
01Weekly Brief
the exploit path, named
02Public Proof
the exposed surface, visible
03Release Gates
the same evidence, enforced
LATEST RESEARCH
What matters this week
Three lines from this week's brief.
02
Operator noteTOOL ABUSE
The dangerous permission lives in the runtime
Shell, fetch, subprocess, and connector paths still turn low-friction prompts into high-impact actions.
Read Archive03
Proof layerPUBLIC EXPOSURE
AI risk gets real when the surface is reachable
OpenClaw shows which endpoints, docs, and agent interfaces are exposed right now.
Open WatchboardTHREAT SURFACE
Where risk becomes real
The surfaces that turn model risk into reachable risk.
CURRENT TRACKING
01Input path
Prompt Injection
Instruction override, hidden tool calls, indirect poisoning, and retrieval chains.
02Runtime
Tool Abuse
Shell execution, downloader chains, connector misuse, and runtime permissions.
03Secrets
Secret Exposure
Leaked API keys, unsafe logs, public config artifacts, and long-lived credentials.
04Exposure
Public Surface
Open docs, unauthenticated endpoints, and exposed agent interfaces.
PROOF LAYER
Public proof beats slides
OpenClaw turns findings into a live watchboard: target, issue, and status.
- Exposure changes the question from can this happen to is this still open.
- Trace risk across docs, endpoints, auth, and runtime behavior.
- Use the same evidence later in ClawPlane and release gates.
OPENCLAW PREVIEWRecent exposed surfaces
TargetRiskIssueStatus
api.assistant-demo.aiHighPublic docs + no authPublic
tools.ops-agent.devHighShell-capable runtimeReview
rag-preview.exampleMediumPrompt injection pathTracked
share.agent-lab.runMediumLeaked token artifactRestricted
INFRASTRUCTURE
Built from the brief
When reading is not enough, the same evidence becomes scanning, policy, and gates.
01
NEWSLETTER
AIPwn Newsletter
Weekly AI security research on exploit paths, exposed surfaces, and ship risk.
Weekly02
CONTROL PLANE
ClawPlane
Policy, scanning, and release decisions for AI systems in one control layer.
Alpha03
SCANNER
ClawScan
Evidence-first scanning across repos, prompts, connectors, and exposed surfaces.
Research-fed04
GATE
ClawGate
Release gates that block risky changes with evidence.
Alpha