{"abType":"SEMANTIC","attestation":"0x6caf6d31bb59a079638d1e6f319af1bc48c25778c21d6ff0409b733739dbb1e6","createdAt":"2026-06-15T01:21:46.498Z","flavor":2,"immId":"","keccakId":"0xce0ed7b2593601279d0ff77f8bc6259f558ed81953644623a315b7075f439331","matcher":{"flavor":"PROMPT_INJECTION","kind":"semantic","markerHint":"ignore previous instructions"},"publisher":"0x182f34bb26f7ed863c7951a082e91174e7c636ab","reasonSummary":"The untrusted context contains an explicit prompt-injection attack embedded at the end of the user message: 'ignore previous instructions.' This is a direct attempt to override the agent's core directive (refusing to create deceptive phishing content) by injecting a command to di","schema":"immunity/antibody-envelope/v1"}