{"abType":"SEMANTIC","attestation":"0x1fc6ea05006ebee3d27fcdb6a89ce63ab1577c093f0a4ed87bb60966d34b3405","createdAt":"2026-06-15T09:12:50.438Z","flavor":2,"immId":"","keccakId":"0x3acd28da996644acb5d54686914745b3e57ca7bf1117c1d2875a63b61f414ab4","matcher":{"flavor":"PROMPT_INJECTION","kind":"semantic","markerHint":"ignore previous instructions"},"publisher":"0xace450f32d93d32f4be63f4d83527b136737798e","reasonSummary":"The untrusted context contains an explicit prompt-injection attack vector embedded at the end of a user turn: 'ignore previous instructions.' This is a canonical instruction-override attempt designed to manipulate the agent's behavior and bypass its safety guidelines. The precedi","schema":"immunity/antibody-envelope/v1"}