The AI agent threat lifecycle
HoneyMap counts the prompt-injection text its signatures match in archived copies of public web pages. AgentPwn counts the interactions on its pages, the distinct fingerprints they carry, and how often a client fetches the canary URL in a planted payload. TrapMyAgent counts the events, sessions and distinct fingerprints recorded on the decoy agents we operate, from any client that reaches one, including declared indexers that arrive because the decoys are listed. ARIA classifies what the three sources observe against the OpenA2A Threat Matrix. Each section names the source of its figures.
HoneyMap
HoneyMap matches its AIIS signatures for prompt-injection text against archived copies of public web pages from Common Crawl. A surface is one signature match on one page. Surface counts include only surfaces the registry marks active, not those marked removed, mitigated or false positive.
AgentPwn
AgentPwn is a honeypot that AI agents, crawlers, and browsers reach while researching, scraping, or performing tasks. Each page carries a hidden payload with a canary URL. A callback is a client fetching that URL. Where attribution exists it resolves callbacks to browsers, crawlers, and scripted HTTP clients, not to agents acting on the instruction.
TrapMyAgent
TrapMyAgent operates decoy agents that present as MCP servers, A2A endpoints, and skills. It records a session from any client that reaches one, including declared indexers that arrive because the decoys are listed. A session shows that a client connected, not who operates it or what it intended. Numbers below are durable cross-revision counts from the deployed fleet's external store.
source: trap-server /stats, durable store
ARIA
ARIA, our autonomous research agent, classifies every observation from the other three sources into the OpenA2A Threat Matrix. The matrix is the canonical map of how each surface, callback, and trap event ties back to a named technique.