Trump's AI accord and Europe's AI Act: two rulebooks, one industry

The document is called the "White House Accord on Superintelligence", subtitled the "Joint Commitment on Frontier Responsibilities". Meta, Google, OpenAI, Anthropic, Nvidia and xAI signed it, and it appeared on Donald Trump's Truth Social account. The Register, which read the text, reported on 30 September that it commits the signatories to nothing enforceable.

Four "layers of controls and audits" are listed. Internal controls over models in cybersecurity, biosecurity and chemical threats. An internal team to check the controls. An independent external auditor. A board committee to receive the reports. The Register notes the text never defines "robust", never says how often auditors inspect, and never sets an evaluation cadence.

Penalties, or the absence of them

German outlet heise online, which also reviewed the one-page paper on 1 October, made the same point in blunter terms. The verbs are "we believe" and "should", not "we will", it said. It added that the first two layers describe processes the AI companies had already announced and in some cases already run, while organisations keep getting breached. The accord's own closing lines leave open the possibility that it "may make sense to codify these steps into laws or regulations".

Compare that with the European Union. The AI Act is legislation, with obligations phased in by risk category, and it is being enforced alongside separate antitrust and data-sharing orders. Google is challenging an EU order to share data with AI chatbot and search rivals, according to reporting summarised by ET Enterprise AI on 29 September. That is a company fighting a binding instrument, not signing a voluntary one.

"We should spend more time making the rules and standards we have work at scale, and less time focusing on fears about theoretical risks or news reports from the Silicon Valley AI doomer bubble."

That is Alexandru Voica, policy lead at UK firm Synthesia, quoted by POLITICO on 24 September in a piece about European tech leaders pushing back on US-generated AI extinction warnings.

Europe's own divisions

Europe is not a single voice either. POLITICO reported on 30 September that the UK's AI minister wants the country to focus on building the technology rather than regulating it. EUobserver reported the same day that EU states are being urged to go beyond the AI Act to protect people from abusive surveillance. Both are in the dossier, and they point in different directions.

Meanwhile the safety incidents that give the debate its urgency keep arriving. MIT Technology Review's 29 September newsletter noted that OpenAI scrapped a model, GPT-6.1 Astra, because it "didn't quite meet the bar", and that Anthropic's IPO filing warns AI could threaten humanity. The same newsletter flagged that Florida has asked a court to block OpenAI from developing new models, and that Anthropic's claimed molecular biology "discovery" has been disputed by biologists who say a pattern it flagged was already known.

The Register's account of the accord notes one further detail. On Tuesday, Trump signed an executive order requiring federal agencies to use the term "Super Intelligence" and the acronym "SI" instead of "AI". The naming changed. The obligations did not.

That asymmetry matters commercially. Omdia's 2027 forecast, published on 30 September, found 59 per cent of organisations expect AI budgets to rise by 10 per cent or more next year, with ROI rather than technical superiority deciding what gets bought. Vendors selling into both markets will need to satisfy Brussels' paperwork and Washington's one-pager. Only one of those has a penalty attached.