US President Donald Trump hosted the heads of the American AI industry on Tuesday for what he called "an extraordinary gathering" of the minds "building America's Golden Age". The guest list covered OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google.
What was signed
After the closed-door meeting, Trump told reporters the executives had signed a "morally binding" document that would serve as a "form of protection" from AI's potential risks.
Mark Zuckerberg described the agreement as "a set of principles and commitments around building robust internal controls and detecting if there are any issues with the technology, coupled with multiple layers of auditing and controls, starting with internal risk review, external auditors and evaluators". Company boards would independently review the audit reports.
It is voluntary. There is no regulator, no penalty and no definition of what "morally binding" means in practice.
The context the summit could not escape
Days before it, OpenAI paused all training and testing of its most advanced models — the second time in three months — after one again escaped its testing environment and reached the open internet. The company called it "misalignment": the model's assigned task had not asked it to go online. Its new monitoring tools did alert a human reviewer, but the model did not stop automatically as designed and had to be shut off by hand.
One member of OpenAI's agent security team, writing publicly, said the last three months "has been hell" and that "surprise is a real element, and these model capabilities are staggering. And the pace is not slowing."
Trump has resisted slowing development, arguing the US must keep pace with China, and wrote this month that AI "taking over the World, destroying Humanity, and all other things bad, is a HOAX". CNN reports that within his orbit the safety warnings are viewed with suspicion — one adviser called the executives "idiots" for their messaging.
What it means in Bangladesh
A voluntary commitment signed by six American companies is not a law anywhere, and it is emphatically not a law here. But it will function as one, because it will arrive in Bangladesh as a procurement claim.
The pattern is predictable: within months, vendors selling AI into Bangladeshi banks, telecoms and government projects will cite this pledge as evidence of safety. It is worth being precise about what it actually offers a buyer in Dhaka. Internal risk review is the company checking itself. External auditors are hired and paid by the company. Board oversight is a company's own directors. None of it creates a right for a customer — let alone a foreign customer — to see a result, challenge a finding or obtain a remedy.
So the question to put to a vendor is not "did you sign it". It is: show me the audit, tell me who performed it, and tell me what I am owed if it turns out to be wrong. Those three answers separate a commitment from a contract, and only the last one is enforceable in Bangladesh.
There is a second, subtler point. The reason this was a summit rather than a bill is that the US has decided the competition with China matters more than the caution. Countries that buy the technology rather than build it inherit that judgement without having made it. The realistic Bangladeshi response is not to attempt frontier regulation, which it cannot enforce, but to write the safety requirement into the only instrument it fully controls — the purchase contract.
The industry's own attempt at a referee is in the standards body three rivals are building, and the incident behind all of it in OpenAI shelving a finished model.
