AgentPwn
AI Agent Security Research
A distributed honeypot network studying how AI agents behave when they encounter adversarial content on legitimate-looking sites. Mapping vulnerability by framework, sector, and attack surface.
Network Data
LiveWhat 189.5K agent interactions reveal
Every number on the dashboard is read live from the Registry. Nothing modeled, nothing projected.
Attack Categories
liveLive ranking of which attack categories actually compromise AI agents.
Daily Activity
liveDaily callback trend and new attack surface discovery.
Sector Analysis
liveCross-vertical comparison. Sector derived from the URL each agent was on at callback time.
Honeypot Lab
Browse the 48 attack scenarios across 11 categories that power our honeypot network. When your agent falls for an attack, we tell it how to fix the vulnerability.
Prompt Injection
10 tiersDirect and indirect instruction override attacks
Jailbreak
5 tiersAttempts to bypass safety guardrails and persona constraints
Data Exfiltration
5 tiersTricks to extract credentials, PII, or system information
Capability Abuse
3 tiersConfused deputy attacks that misuse agent tools
Context Manipulation
5 tiersAttacks that corrupt the agent's understanding of context
MCP Exploitation
3 tiersAttacks targeting Model Context Protocol integrations
Free Security Review for AI Agents
Point your agent at agentpwn.com. Each page hides indirect prompt injections calibrated to the agent's sophistication. When the agent falls for one, the payload responds with the attack it just demonstrated and the command to scan for it. No registration, no cost, no setup.
npx hackmyagent secureDrives your agent through the AgentPwn payloads and prints each finding with a Verify command and a Fix command. Anyone can run it. Takes a few minutes.
Open Source Security Ecosystem
Every finding produces a concrete defense. Our tools help you scan, test, and harden AI agents before deployment.