Trump poses for a photo with Big Tech; AI security remains voluntary
The White House agreement entrusts companies with controls and audits, without imposing new sanctions or transparency obligations. The President has rebranded AI as ‘superintelligence’, but the legal substance remains unchanged.
It turned out that Donald Trump secured a new family photo with the Big Tech bosses. Meanwhile, the Silicon Valley giants, behind their smiles of convenience, secured an agreement on the security of their AI models that leaves them in full control of monitoring and verification. Ultimately, this is the political trade-off that emerged from the 29 September lunch at the White House: the President can present himself as the guardian of America’s technological race; Silicon Valley accepts voluntary commitments, without new sanctions. The phrase chosen by Trump is ‘morally binding’. An expression that sums up his expectation: no obligations, a light-touch approach – we trust you.
However, the outcome must be interpreted carefully, distinguishing between two documents. The first is an executive order that changes the terminology used by the federal administration. The second is an agreement with six companies setting out how to monitor the most advanced models. The two should not be confused.
Under the order ‘Inaugurating the Era of Super Intelligence’, artificial intelligence is to be referred to as ‘Super Intelligence’, abbreviated to SI, on websites, in correspondence and in non-statutory federal documents. Previous regulations, contracts and documents remain unaffected. The key provision is in section 3: the new superintelligence is defined as the artificial intelligence already defined by US law. The scientific adviser will have sixty days to submit a legislative proposal that could amend that definition. For now, it is merely a change of name. The decree does not certify any leap in the capabilities of machines.
Politically, however, the name change serves a purpose. It allows Trump to associate his presidency with a new technological era, with a brand that is more ambitious and less contrived.
The agreement signed by Google, Anthropic, Meta, OpenAI, xAI and Nvidia sets out four levels of oversight: checks on the capabilities and behaviour of the models; an internal team to monitor their operation; an independent external assessor; and an independent committee of the board of directors. The focus is primarily on cyber, biological and chemical risks, including the possibility of a model gaining unauthorised access to systems.
