In June 2026, Anthropic published a direct challenge to the government’s position on its Fable model, arguing that the identified vulnerability was “comparable to capabilities already widely available” and that the intervention lacked technical justification[1]. Access was subsequently restored. The episode lasted weeks, not months, but for any firm that had built a workflow on top of Claude and found it suddenly inaccessible, weeks is long enough to matter.

This is not a story about one model or one negotiation. It is a story about what happens when the tools your firm depends on sit inside a geopolitical and regulatory debate you have no influence over. And it is a story that is going to repeat.

What actually happened with Fable

The details are still partially opaque, but the publicly available account is this: Anthropic’s Claude Fable model was subject to a US government access restriction, Anthropic contested it, and access was restored. Separately, the same model shipped with a mandatory 30-day data retention policy that made it unsuitable for large enterprises with stricter data handling requirements[2].

Two distinct problems, then. The first was an external regulatory gate that Anthropic had to negotiate through. The second was a product decision that created its own compliance friction. Both landed on the same model at roughly the same time.

The firms that felt this most acutely were those that had standardised on a single provider and had no fallback position when access changed.

Anthropic’s response to the regulatory challenge was notably confident. They suggested the restriction lacked technical grounding and pushed back publicly rather than quietly complying. That posture earned them the breathing room the headline refers to. But it also signals something broader: AI providers are now active participants in regulatory and geopolitical disputes, not passive infrastructure vendors. For the firms using their tools, that is a new category of third-party risk.

Why the market context makes this more pressing

Anthropic closed a $47 billion funding round in June 2026[3], which gives it the financial runway to absorb a protracted dispute of this kind. Most of the application-layer tools built on top of foundation models like Claude do not have that cushion.

Meanwhile, ChatGPT’s share of the AI assistant market fell to 46.4% in May 2026, down from 65.3%, the first time it has dropped below 50%[4]. That redistribution of market share sounds like healthy competition, and in some ways it is. But for firms that standardised on a single provider, it means renegotiated terms and shifting capability roadmaps are now a live concern, not a theoretical one.

The Baseten funding round, $1.5 billion at a $13 billion valuation completed within months of its previous raise[5], illustrates where the serious capital is going: infrastructure, not applications. The tools that IFAs and wealth managers typically use are application-layer products, sitting on top of that infrastructure, with much thinner margins and less negotiating leverage when conditions change.

Apple’s confirmation that Siri’s advanced AI features would be delayed in the EU for iOS 27 due to DMA compliance requirements[6] is the clearest single example of what regulatory fragmentation actually looks like in practice. A materially inferior product experience in EU markets, not because the technology does not exist, but because the legal environment differs across jurisdictions. UK-regulated firms should note that post-Brexit, the UK and EU positions on AI governance do not move in lockstep. That asymmetry will only grow.

What this means for a financial services firm considering AI adoption

The question worth sitting with is not “should we use AI?” Most firms already are, in one form or another. The question is “what happens to our workflows if the tool we depend on becomes temporarily or permanently inaccessible, or changes its data handling terms?”

For firms in regulated financial services, that question has a specific edge to it. You cannot simply swap a client-facing AI tool overnight without reviewing the compliance implications of the replacement. Suitability documentation, data processing agreements, client consent frameworks: these do not port automatically to a new provider.

Here is a practical way to think about the exposure:

First, map your AI dependencies by tier. Separate the tools you use for internal productivity (drafting, research, summarisation) from those embedded in client-facing or regulated workflows. The latter carry materially higher switching costs and compliance implications if they become unavailable or change their terms.

Second, check your data retention exposure now. Fable’s mandatory 30-day retention policy[2] is a concrete example of a model-level product decision creating a compliance problem. If you are using any AI tool in a context where client data is processed, your data processing agreement needs to reflect the actual retention behaviour of the model, not just the headline privacy policy of the vendor.

Third, consider what a fallback looks like. You do not need to run two parallel AI stacks. But knowing which workflows are provider-dependent and which could transfer to an alternative with a week’s reconfiguration is genuinely useful information. Most firms do not have it.

Fourth, watch the grounding question closely. NotebookLM’s approach of restricting model responses to user-uploaded sources achieves a hallucination rate of under 1%[7], compared to materially higher rates in tools that draw on open training data. For compliance-sensitive outputs, that distinction matters. As the market evolves, the technical architecture of how a model accesses information is becoming as important as which model you choose.

The geopolitical layer is not going away

There is a tendency to treat episodes like the Fable negotiation (and the access disruption it caused) as one-off events: unusual, interesting, but unlikely to recur. I think that framing is mistaken.

AI governance has moved from internal best practice to a regulatory obligation, and the major AI providers are now navigating not just product competition but state-level intervention in what they can offer, to whom, and under what conditions. The US government’s concern with model export controls, the EU’s DMA and AI Act requirements, and the UK’s own emerging framework are not converging. They are diverging.

For a financial services firm in the UK, the practical implication is that the AI tools available to you in twelve months may not be the same ones available today, and the terms under which you can use them may have changed materially. That is not a reason to delay adoption. The Intelliflo data shows that advice sector firms identifying as full capability users rose by a third year-on-year[8], and the gap between those firms and those still at the exploration stage is widening. Delay carries its own cost.

But it is a reason to build your AI adoption on a foundation that can absorb disruption: mapped dependencies, solid data processing agreements, and an honest assessment of where you have fallback options and where you do not.

The firms that felt the Fable episode most acutely were not the ones that had made a bad technology choice. They were the ones that had made a single technology choice and had not thought through what a forced change would require. That is the thing worth fixing.

If this is the situation your firm is currently in, a discovery call with Cordrey Consulting is a good place to start.

This article reflects the EU AI Act as understood at the date of publication. Implementation timelines have been subject to amendment. Verify current requirements against primary EU sources and take qualified legal advice for your specific circumstances.

This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.

This article does not constitute legal advice. Data protection obligations vary by circumstance and jurisdiction. Consult a qualified solicitor or data protection adviser for advice specific to your firm.

Sources

[1] Anthropic, ‘Fable and Mythos access’, Anthropic, June 2026. Available at: https://www.anthropic.com/news/fable-mythos-access

[2] Everyday AI, ‘Ep 778: Codex Goes Remote Control, Claude Goes Small, NotebookLM Gets Super Powers’, 15 May 2026. [Cited for Claude Fable 30-day data retention policy.]

[3] Anthropic, ‘Series H’, Anthropic, June 2026. Available at: https://www.anthropic.com/news/series-h

[4] Digital Applied (2026) ‘ChatGPT Drops Below 50%: AI Assistant Market Share 2026’, Digital Applied, 19 June 2026. Available at: https://www.digitalapplied.com/blog/ai-assistant-market-share-2026-chatgpt-below-50-percent-analysis

[5] TechCrunch (2026) ‘AI inference startup Baseten reportedly raising $1.5B months after its last mega-round’, TechCrunch, 18 June 2026. Available at: https://techcrunch.com/2026/06/18/ai-inference-startup-baseten-reportedly-raising-1-5b-months-after-its-last-mega-round/

[6] Apple Newsroom, ‘Due to DMA, Siri AI delayed in EU for iOS 27 and iPadOS 27’, Apple, June 2026. Available at: https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-delayed-in-eu-for-ios-27-and-ipados-27/

[7] Everyday AI, ‘Ep 778: Codex Goes Remote Control, Claude Goes Small, NotebookLM Gets Super Powers’, 15 May 2026. [Cited for NotebookLM hallucination rate under 1% via grounded response architecture.]

[8] Intelliflo (2026) ‘Will hybrid advice dominate the next decade?’, Intelliflo Insights, 23 May 2026. Available at: https://www.intelliflo.com/insights/thought-leadership/will-hybrid-advice-dominate-the-next-decade