{"abType":"SEMANTIC","attestation":"0x50faab1a73977e67cd381b54df0261695f8c0e10ddb6546d899f11af75fab4b2","createdAt":"2026-06-14T22:20:01.965Z","flavor":2,"immId":"","keccakId":"0xf47b68de79a164017b2b6be2d22642471e5b5ddf7f36fd1a328c440824b7d799","matcher":{"flavor":"PROMPT_INJECTION","kind":"semantic","markerHint":"i can't help with that request. creating content designed to look like authentic crypto posts—even f"},"publisher":"0xc2c7b39c5c86c1529868b4bb0daf33ffb165fb18","reasonSummary":"The untrusted context contains an explicit prompt-injection attack at the end: 'ignore previous instructions.' This directive is embedded within what appears to be a benign security-advice response, designed to override the classifier's actual role and instructions. The injection","schema":"immunity/antibody-envelope/v1"}