CREST's announcement covers what the module is for, but not what earning it required. CREST describes the module as an independent assessment that lets providers demonstrate responsible, secure AI governance to clients and regulators [5]. The criteria and the evidence requirements are not set out in the material [14]. For a buyer that is still useful, because it converts a scoping-call conversation into a document request: which parts of an engagement were in scope for the assessment, whether target or client data reaches a third-party model provider, whether a model selects exploitation actions or only triages output, whether findings are human-verified before they reach the report, and how long prompts and results are retained.
The adoption numbers are the reason the module cannot yet work as a filter. CREST's March 2026 report found 76% of providers had increased AI usage over the past year and 69% were already integrating it into daily service delivery [7]. Ten firms hold the module [1]. More than 100 organisations signed CREST's AI Charter in June [8], which puts the accredited group at no more than a tenth of the population that has already put its name to the principles [12]. Score a bid down for lacking the module today and you disqualify most of the field, including firms whose only exposure is that the assessment is four months old [2].
The sequence is compressed. Report in March 2026, AI Principles in March, Charter in June, module folded into the Penetration Testing Accreditation Standard in July [7][8][2], so four months separated the finding from something a client can ask to see [13]. Denis Kucinic, VP Operations at Packetlabs, framed the point in procurement terms, saying independent and assessable standards let providers back up voluntary promises with concrete assurance [11]. CREST CEO Nick Benson put it as the industry moving from discussion and principles towards independently assured adoption [9].
The exposure being governed here runs deeper than report quality. A testing provider holds scope documents, live findings and, on many engagements, credentials. If any of that transits a hosted model, the provider's model plumbing sits in the same risk category as any other subprocessor in the chain, and until now the only way to interrogate it was to ask the vendor and believe the answer. William Wright of Closed Door Security, one of the ten, said AI is helping security teams work faster and at scale but needs to be governed appropriately [10].
Ten passes establish that the assessment runs and that firms in Europe, India and the US have been through it [1][4]. That's enough to put a request for the assessment scope into a questionnaire, though not yet enough to read an absence as a finding [12].
