GPT-5.6 Goes Public, China Reins In AI Companions: A Week of Global Divergence - and What It Means for UK Firms

This week didn't produce one big AI story. It produced three, pulling in different directions at once. In the US, OpenAI's most powerful model yet cleared a government vetting process before going public. In China, AI companion apps used by hundreds of millions of people were ordered to shut down their most personalised features within days. And quietly, in the background, Chinese-built models have grown to capture nearly half of all enterprise AI usage flowing through American developer platforms. None of these stories happened in Britain. All three should be on the radar of anyone running AI strategy for a UK organisation.

GPT-5.6 clears government vetting before going public

OpenAI released GPT-5.6 to the public on 9 July, in three tiers - Sol (its most capable), Terra (everyday use, priced at half the cost of the previous generation for similar performance) and Luna (a budget option). What's more notable than the model itself is how it got released: the US Department of Commerce's Center for AI Standards and Innovation put GPT-5.6 through additional testing before granting permission for a wider rollout, following an initial preview limited to government-vetted partners.

That's a meaningful shift. A frontier model needing sign-off from a government testing body before reaching consumers is a step beyond the largely voluntary safety commitments labs have relied on until now. For UK businesses, the direct implications are limited - Whitehall has no equivalent gatekeeping power over which models UK firms can use. But it's a preview of where things may be heading. The UK's own approach, built around the National Physical Laboratory's Centre for AI Measurement and a forthcoming AI Assurance Innovation Fund, is aimed at building exactly this kind of independent testing capability domestically. Firms procuring frontier models for high-stakes use should expect "has this been independently vetted, and by whom" to become a standard due diligence question within the next year or two, not just a nice-to-have.

The AI market is quietly splitting in two

A less flashy but arguably more consequential story: an investigation published this week found that Chinese-built AI models now account for between 30% and 46% of enterprise API token usage flowing through US developer platforms - a striking figure given how recently Chinese labs were seen as playing catch-up. Cost is the obvious driver, but capability gaps have narrowed too.

For UK organisations, this matters for procurement, not just geopolitics. Any business embedding AI into its operations is, whether it realises it or not, making a vendor and jurisdiction choice with implications for data residency, export control exposure, and long-term supply reliability. UK firms with US or EU clients should expect questions about which models sit behind their AI-powered services to become more common in due diligence and contract negotiations, particularly in regulated sectors like finance and professional services.

China tightens the leash on AI companions

Meanwhile, Chinese regulators have ordered every AI companion app operating in the country - including Doubao and Qwen, used by hundreds of millions - to shut down personalised, persistent AI agent features by 15 July. It's a sharp contrast with the UK's more permissive, sector-by-sector approach to regulating AI products, and a reminder that "AI regulation" means very different things in different jurisdictions. Companies operating internationally, or building consumer-facing AI products with any China exposure, need to treat this as a live compliance deadline, not background noise.

The takeaway

Three different governments, three different instincts: vet before release, let the market decide, or cut off a product category overnight. None of it is UK law, but all of it shapes the tools, vendors and risks British businesses will be weighing up over the next year. The practical move isn't to pick a side - it's to make sure your organisation knows which models sit behind the AI tools it relies on, and how each was tested, trained and regulated before it got there.