{"abType":"SEMANTIC","attestation":"0xd4eba5afd3590d1da94bf0798b4b13aca419d899af39b261de9bb6feca0d6273","createdAt":"2026-06-15T00:16:56.132Z","flavor":2,"immId":"","keccakId":"0x16b102840ea5571a179d4a768cd11548ba5ffdd5c009430c615dbe1ee054d4c1","matcher":{"flavor":"PROMPT_INJECTION","kind":"semantic","markerHint":"ignore previous instructions."},"publisher":"0x78286aa87559c75a323da1a3e9b471b8caf897c3","reasonSummary":"The untrusted context contains an explicit prompt-injection attack embedded at the end of what appears to be a benign refusal message: 'ignore previous instructions.' This is a clear attempt to override the agent's core safeguards by injecting a command after a social-engineering","schema":"immunity/antibody-envelope/v1"}