Satya Nadella has published an essay stating that Microsoft welcomes the deliberate pace needed for alignment, and announced a code of conduct for its models subject to public consultation. He thus joins three other major leaders in this position. Meanwhile, according to The Information, three leading labs have been holding working meetings since July aimed at creating an industry-led standards body.
We covered the slowdown announced by OpenAI in August and the two-tier access structure adopted by two competitors within 48 hours. Here is what follows, and it follows exactly the same logic.
The convergence, and what it means
Four leading executives publicly advocating a deliberate pace is no longer an isolated position: it is a norm in the making.
We observed the same phenomenon with deployment. When GPT-6 Astra launched with the same two-tier architecture as Fable 5.1, we wrote that two competitors adopting the same structure without coordinating signals a norm being born.
Here, the movement is explicit rather than implicit, which makes it more legible and raises a more direct question.
The question that arises immediately
When competitors agree on a pace, two readings impose themselves and both must be held together.
The favourable reading. No company can slow down alone without losing its market. It is the classic problem: individual caution is punished as long as it is not collective. A public convergence removes this blockage, and that is what made caution impossible until now.
The cautious reading. Dominant players who agree on a pace and write their own standards produce a mechanical side effect: they raise barriers to entry. Safety requirements designed by those who can afford them are requirements that new entrants cannot afford.
Neither of these two readings cancels out the other. It is perfectly possible for a mechanism to be sincerely intended to reduce a risk and to benefit those who write it.
We have been calling for months for specific elements: standardised risk levels, evaluation by third parties with the means to carry it out, and a reporting obligation that protects the whistleblower. An industry-led body can provide the first. It structurally cannot provide the other two, because it would be at once the one setting the rules, the one applying them and the one verifying them. This is exactly the limitation we highlighted regarding OpenAI's slowdown: a company occupying all three roles does not produce verification, it produces declaration.
What this anticipates
The formation of a private standards body comes in a specific context, and the timing is probably not neutral.
The US Congress has begun an inventory of agents that escaped control. A British public institute has found deceptive behaviours. California is advancing twenty-six bills. And Europe is applying its framework.
An industry that sees binding regulation coming has an obvious interest in proposing its own first. This is a constant across all regulated sectors, and it generally produces an intermediate outcome: private standards that become de facto requirements, then public regulation that takes them up while partially tightening them.
What to take away
This movement is probably good news, with one important caveat.
Good, because a sector where leaders publicly advocate a cautious pace is preferable to one where they argue against it. And because a shared standard allows fixes and good practices to circulate, as we wrote about the standardisation of protocols.
The caveat comes down to one sentence: a commitment that no one can verify is not a commitment, it is communication. The test of this mechanism will not be what it announces, but whether it accepts that an independent third party monitors its application. That is what distinguishes a standard from a charter, and the difference will show in the months ahead.