The restriction on AI tools is the single loudest frustration in every discovery workshop we run, in every function, and it is the one decision standing between most businesses and their fastest, highest-value AI opportunities. The concern behind it is legitimate. The block itself is the wrong answer, because it does not do what it was meant to do.

What the block is for

Proprietary production data, process parameters, pricing, customer records and source code should not be pasted into a consumer-grade AI tool whose terms allow the provider to train on it. That is a real risk, and the IT and legal functions that raised it were right to. Any argument for opening access has to answer it directly rather than wave it away.

What the block actually does

It displaces AI use rather than preventing it. In workshop after workshop, staff describe using personal ChatGPT and Claude accounts “outside work” to get the job done: drafting, summarising, translating, writing code, and in at least one case asking a general model to make judgement calls on safety standards. The business has no visibility of any of it, no control over what is shared, no record of what was generated, and no training in how to use the tools well. That is precisely the data-leakage risk the block was meant to prevent, made worse, because it is now invisible.

The recurring phrase from participants, in their own words, is that the blocker is licensing, not data.

What the block costs

The two cheapest, highest-value opportunities in almost every programme sit at the base of the Opportunity Pyramid: a sanctioned productivity assistant for the routine work every function describes the same way, and content and creative tooling that runs largely on software the business already owns. In our most recent full programme, twenty-six separate ideas raised across five functions consolidated into the first of those, and both were ready to switch on. Both were blocked only by access. The block trades a controllable risk for an uncontrollable one and pays for it in forfeited value.

The alternative: governed access

The controls are well established and none of them is exotic.

  • Enterprise licences with no-training terms. The enterprise tiers of Microsoft 365 Copilot, ChatGPT Enterprise and Claude for Work or Enterprise contractually commit that business data submitted through them is not used to train the underlying models. Confirm the exact terms at contracting; they answer the concern that drove the block.
  • Data classification and data-loss prevention. Clear rules on what may and may not be shared, backed by DLP controls, so sensitive material is protected whatever the tool.
  • A sanctioned-tool list and phased, role-based access. Extend whatever pilot exists into a governed rollout, open access by role and use case rather than all at once, and give people an approval route for new tools so the list stays current.
  • A usage policy and AI literacy. So staff know how to use the tools safely. This is also now a regulatory requirement under Article 4 of the EU AI Act, in force since February 2025.

Governed access manages the risk better than the block does, because the use becomes visible, trained and contractually protected, and it unlocks the fastest value in the programme at the same time.

What to say to the people who imposed the block

That they were right about the risk and that there is now a better way to manage it. The sanctioned route gives them visibility and control they do not currently have; the block gives them neither. Enforcement without an alternative only drives use further underground, and the staff using personal accounts are evidence of demand, not of misconduct.

The governed-access model is part of our AI governance service, and the rollout that follows it is under AI tools and platform rollout.