Outside-Counsel Memo Cross-Examination Pack for In-House GC

For in-house General Counsel at 50-500 employee EU/UK companies whose product touches AI Act applicability questions. The GC enters discrete claims from an outside-counsel memo (article cited, classification reached, carve-out invoked, 1-3 sentence reasoning chain) into a structured form. AE runs adversarial multi-model debate against the AI Act text, AI Office FAQs, and ENISA bulletins, then returns named challenge questions in legal vernacular ('the Art. 6(2) carve-out rests on an unstated premise about purpose limitation; please defend or narrow'). The GC sends those challenges back to counsel. The primary graded resolution event is counsel's response within 7-14 days: concession, refinement, or substantive defense — a fast, objective evaluator-side signal that maps directly onto AE's 6-pattern autopsy. Internal pattern names (Fatal Grounding Immunity, Concession Laundering) drive the engine and the GC's confidence ledger but are translated out before contact with counsel. Subsequent regulator publications are a slow secondary signal, not the primary truth path. Recurs on every feature, market, or model change through 2027+.

Under-resourced mid-market EU/UK GCs need a cheap, structured way to cross-examine outside-counsel AI Act memos, and AE's adversarial debate plus 7-14 day counsel-response grading is the rare legal-AI shape with a fast objective signal.

Killed: fatal_objection_both_confirm.