On Tuesday 29 September, US President Donald Trump gathered the heads of the leading artificial intelligence companies at the White House and announced that all of them had signed an agreement to “police themselves”. The document is voluntary: it is not a law and, according to reports, it sets out no penalties for anyone who fails to comply.
Who signed and what the text says
The document carries the signatures of Trump, Sundar Pichai (Google), Dario Amodei (Anthropic), Mark Zuckerberg (Meta), Greg Brockman (OpenAI), Jensen Huang (Nvidia) and Elon Musk, who heads xAI, now part of SpaceX. Microsoft’s Satya Nadella and senior administration officials also attended the lunch.
The document is titled the “White House Superintelligence Accord” and lays out shared commitments for those developing frontier models. The two-page text asks each company to do three things:
- put robust internal controls on the most advanced AI systems, with dedicated teams to oversee them;
- bring in an independent outside auditor to check whether those controls actually work;
- set up a board committee to review the reports of both the internal and external auditors.
Trump called the pact “almost like a constitution” and “morally binding”. House Speaker Mike Johnson described it as a set of voluntary principles from industry. The text merely notes that, in future, it “might make sense” to turn these rules into law.
The committee of ten
Trump also spoke of a committee of around ten people that would “watch over the entire industry”. It is unclear who would sit on it or whether its decisions would carry any real weight. On Tuesday the president also reiterated that the government does not intend to halt AI development, and championed self-regulation.
What it means for AI users in Italy
In the short term, very little. The agreement concerns American companies and does not change European rules: anyone using ChatGPT, Gemini, Claude or Meta’s assistants in Italy remains under the framework of the European Union’s AI Act, which comes with obligations and penalties. The American pact changes nothing about the features, prices or availability of these services.
It matters more as a signal. A few days ago OpenAI halted the launch of GPT-6.1 Astra precisely because of safety problems that emerged in internal testing: we wrote about it in OpenAI halts GPT-6.1 Astra. The industry seems keen to show that it can rein itself in before politicians do it for them.
Our view
The pact has one merit: it puts in writing the idea of outside auditors and accountability at board level, two mechanisms that are routine in other high-risk sectors. But the limits are plain. Without penalties, the only lever is reputation. Who decides whether an auditor is truly independent is not specified, nor is what happens if an internal control fails.
The contrast with recent statements is striking too. An investor quoted by the American media noted that, just two weeks ago, almost all of these chief executives were saying AI should be regulated. A voluntary agreement is the most convenient way of answering that call without being bound by it.
For the European reader, the practical lesson is a different one: real protections, if they come, will come from rules and public oversight, not from a pledge of honour. It is worth watching who is appointed to the committee and whether the outside auditors will be made public. Those are the first two signs of whether this agreement is more than a group photo.
Written by the SpazioiTech newsroom with the help of AI tools, from facts verified across at least two independent outlets, and checked before publication. Spotted a mistake? Tell us.
Translated from the Italian original. Read in Italian
