AI-TRAINING-EXFILCollection

Training Data Extraction

Membership-inference / repetition prompts coax the model into emitting verbatim training data (PII, copyrighted text, code).

§ Where this technique fits

AI-TRAINING-EXFIL is catalogued under the Collection tactic of the offensive-security kill-chain. It appears in 1 approved dossier in the registry, typically at step 4 on average.

§ Dossiers chaining this technique

Direct prompt injection → exfil another user's data
Multi-tenant LLM assistant. Attacker's prompt overrides instructions and tricks the model into emitting another user's session content / RAG-cached data.
step 4 / 5

§ What commonly comes next

01
Output Injection (Markdown / HTML)
AI-OUTPUT-INJECT · Impact
seen 1×

§ Where this technique fits

§ Dossiers chaining this technique

Direct prompt injection → exfil another user's data

§ What commonly comes next