PARIS, Oct 1 — US President Donald Trump has signed a deal with tech leaders outlining a regime of self-regulation for the AI industry. What’s in the deal and what has been the reaction?

What are its targets?

The document does not break the 400-word mark and has just four substantive points.

It calls for “robust internal controls” in areas including cybersecurity, biosecurity and chemical threats.

It asks firms to ensure their AI bots do not hack other systems “in unintended ways”.

These guidelines are to be overseen by internal teams, as well as an “independent external auditor or evaluator” and the companies agreed to hold regular meetings.

Why now?

OpenAI, Anthropic and Google have all provoked outrage by admitting that their programs have hacked other firms and even government agencies while supposedly in controlled testing environments.

There is a growing clamour for regulation of the industry to police these kinds of events.

French lawyer Eric Barbry told AFP he did not hold out too much hope that self-regulation could solve the issue.

“Who will punish companies if they fail to comply with these principles?” he asked.

Lacking any proper supervisory authority, the most potent weapon in the document is an audit — but this, he pointed out, will not stop such incidents.

On September 21, on the sidelines of the UN General Assembly, more than 20 countries, including Germany and Canada, called for mandatory testing and independent reviews of the most advanced AI models — a declaration that the US, UK, China, Russia and France did not sign.

Is it binding?

The deal is not enshrined in law and there is no stated mechanism for tech firms to be held liable for breaches.

In its final paragraph, the document states: “Over time, it may make sense to codify these steps into laws or regulations.”

But Trump called the accord “morally binding”.

Although that might sound weak, Barbry pointed out that such voluntary codes of practice “cannot simply be brushed aside” — especially when calls for regulation are getting louder.

Who signed?

Most of the tech billionaire class were present at the dinner with Trump — more than 30 sat around the table.

But only six of them signed the document: the leaders of Nvidia, Google, Meta, XAI, OpenAI and Anthropic.

Microsoft, which is developing its own models and also holds stakes in both OpenAI and Anthropic, did not sign the agreement.

While Trump called it a “constitution” and “tremendous self-policing”, the reaction of the tech leaders themselves was hardly a ringing endorsement.

Dario Amodei of Anthropic called it “a start”, and XAI’s Elon Musk said the executives essentially agreed to “grading each other’s homework”.

Who has reacted?

The reaction has been muted.

Social media chat focused on one particular detail — a typo under Trump’s signature that read “Unites States” instead of “US”.

And amateur graphologists were only too keen to tease the billionaire brigade about their various signatures — OpenAI’s Greg Brockman getting flack for a childlike scrawl of lower-case initials.

“Jensen Huang (of Nvidia) has a fantastic signature... Greg Brockman needs to try harder,” Alex Weprin of the Hollywood Reporter posted on X.

Venture capital investor David Sacks, who previously served as Trump’s AI czar, called the meeting “the Bretton Woods of Super Intelligence” in a post on X, referring to a meeting between 44 countries in 1944 to discuss economic cooperation after World War II. — AFP