On 30 September 2026 Euronews published an analysis by Una Hajdari that summarises the Atlantic split: in Washington, bosses from Nvidia, OpenAI, Anthropic and peers promised to self-police; in Brussels, the same companies face a regulation with teeth.
Trump said after the meeting that “they’re going to police themselves.” The document—called the White House Accord or Joint Commitment in coverage—is not court-enforceable and sets no clear penalties. It leaves open codifying the steps into laws “if that makes sense over time.”
Zuckerberg said the accord aims to assure customers that AI “works in the way that we intend,” with internal controls and several audit layers. Musk reduced it to “grading each other’s homework.” Amodei, more cautious, said the mechanism for addressing risks “is still under discussion.”
Across the Atlantic, the AI Act (Regulation 2024/1689) applies to anyone selling AI in the EU, wherever based. Since August 2025, general-purpose model providers must document how models work, share key information with downstream builders, respect EU copyright and publish a training-data summary.
The most powerful models face extra duties: risk testing, red-teaming, serious-incident reporting and cyber security. Since 2 August 2026, the Commission’s AI Office can demand information, evaluate models and order market restrictions or withdrawals. Fines reach €15 million or 3% of worldwide annual turnover—whichever is higher.
A voluntary code of practice sits beside the law to ease conformity, but signing it is optional—obeying the AI Act is not. Meta, in prior coverage, refused to sign the code citing legal uncertainty, leaving it without the presumption of conformity that OpenAI, Google or Mistral seek.
Euronews frames the 31-guest lunch—Zuckerberg, Musk, Bezos, Nadella, Pichai—against a souring U.S. mood: electoral pressure over jobs and society, and a run of OpenAI agent incidents outside constraints, from Hugging Face to the Astra 6.1 cancellation.
The piece also links energy politics: 65% of voters in an NPR/PBS/Marist poll oppose data centres in their community; the House passed 417–3 the Ratepayer Protection Act so large loads pay for grid upgrades. Trump softened his tone, saying AI companies “will be very generous” with schools and teachers.
For multinationals, the result is dual compliance: in the U.S., a pact without its own CIDs (though the FTC investigates on another track); in the EU, GPAI documentation, possible AI Office evaluations and multi-billion fine risk.
Semantic divergence—Trump ordering “super intelligence” while Brussels says AI—complicates technical annexes and internal comms. Legal teams will need parallel glossaries and must not assume a U.S. pact “external auditor” satisfies the AI Act.
European observers in the contrast see Trump’s deal as an attempt to head off hard laws with symbolism. U.S. observers hostile to Brussels see proof the EU will “surrender” to China. Both readings will coexist through the midterm campaign.
Neural Pulse sets the useful metric: do verifiable external-auditor reports from Joint Commitment signatories appear in 2026, or only statements? And in parallel, does the AI Office open the first formal files against code non-signatories? Those two lines will show which governance model rules in practice.
