Annex IV Interrogation Pack for Medical-Device SMEs Buying AI Components

A subscription challenge-pack product for quality and regulatory leads at small UK/EU medical-device manufacturers integrating AI/ML components from external suppliers. Each pack maps to EU AI Act Articles 9-15 (risk management, data governance, robustness) and produces a structured set of notified-body-style probes the buyer puts to the vendor before signing a component contract. The buyer pastes the vendor's attestation snippets into the pack manually; AE returns a structured evidence-gap response plus a classification of the vendor's rhetorical pattern under AE's 6-pattern autopsy taxonomy — concession laundering, epistemological shielding, premise-conclusion severing — none of which generic AI Act tooling detects. With the Omnibus trilogue collapsed and Aug 2 2026 high-risk obligations now legally binding, medical-device SMEs that integrate a non-conformant AI component face full re-conformity + €35M penalty exposure. AE's adversarial multi-model debate is uniquely suited to generating the probes; AE's autopsy taxonomy is uniquely suited to naming the evasive patterns vendors deploy under regulatory pressure.

AE's adversarial debate and rhetorical-pattern autopsy give medical-device SME buyers a uniquely defensible way to interrogate AI vendor attestations under the post-Aug-2026 MDR + AI Act liability regime.

Engine verdict: GATHER_MORE_SIGNAL (WORTH_SKIMMING). Real regulatory pain, real differentiation — but buyer leverage, ICP coherence, and domain-transfer of AE's taxonomy are all unverified.