Sam Altman Says Accept Some Bad Things. Yours, Ideally.
Altman wants the world to absorb a few unspecified harms for the greater good. The White House's new oversight body has no budget, staff or deadline.
Sam Altman has a proposal for you. "We believe," he told POLITICO Decoded this week, "that the world should accept some bad things happening for the benefits of this technology and people having the agency." Read that sentence twice and it gets worse each time. "Some bad things happening" is a grammatical masterstroke: no subject, no object, no victim, no hospital, no regulator, no invoice — just weather. And the sentence does not land on the harm. It lands on the upside, "people having the agency," which is precisely what the person on the receiving end of the bad thing will not have.
Note the *we*. OpenAI's revenue is not a person and its board is not a jury. In practice, "we believe the world should accept some bad things" parses as: we believe other people should accept them. It is a liability waiver written in the first person plural, covering events that will occur in the third.
In the same week, the White House unveiled its institutional answer: the Super Intelligence Force. Five days ago I wrote that the only binding clause in the President's AI event was the rebranding — that "super intelligence" was a marketing term wearing a policy hat. It now has an org chart. The Force will be led by the Director of National Intelligence, Jay Clayton, alongside the FTC's Andrew Ferguson and the Pentagon's Emil Michael, which is to say intelligence, competition and procurement: three briefs that already exist, now sharing a letterhead. There is no budget, no start date, no deadline, no staff count and no reporting requirement, and nobody has said whether the thing will produce recommendations, draft legislation, or a commemorative coin. It is a force in the sense that gravity is a force.
Malaysia, meanwhile, is doing the boring thing. Digital Minister Gobind Singh Deo stood up in parliament on Monday and said his government is drafting the country's first dedicated AI statute — risk-based, enforcement to follow, early 2027. There is no draft text and no consultation timetable, so this is a direction of travel, not a triumph. But a minister saying "we are writing a law" in a chamber where he can be questioned about it is a categorically different act from a press release naming three men and no numbers. The country with the frontier labs has produced a task force.
This is not for want of public appetite. Stagwell's Harris Poll and HarrisX, surveying for the Harvard Center for American Political Studies, found 74% of US voters want a bipartisan agency to regulate AI. Pramila Jayapal's framework would do something close to that — national charters, government testing facilities, a kill switch — and, as I wrote on Friday, it has no bill number, no committee referral, and roughly no chance. A framework, remember, is a document explaining what a law would contain if anyone intended to pass one. The demand is real. The supply is décor.
The trouble with "accept some bad things" is that it is never the speaker who does the accepting. The bad things that actually arrive are unglamorous: a cloned voice that empties a pension, a triage tool that downgrades a chest pain, a hiring model that quietly filters out a postcode. These are not hypotheticals waiting for us at the frontier; they are the present, distributed in small unlit increments, which is exactly why "some bad things" is the right phrase and "which bad things, happening to whom, compensated how" is the question the sentence exists to dodge. Masayoshi Son, who has put $64.6bn into OpenAI, stood up in Kyoto this week and said superintelligence in the wrong hands could be "super dangerous." Everyone in this industry agrees the risk is real. Nobody in it agrees the bill is theirs.
If Altman genuinely believes in the trade he is offering, he should price it. Not a blog post — a line item. A public incident register. A compensation pool funded from the revenue this technology generates. Mandatory disclosure of harms inside a fixed number of days, with the counting done by someone who does not report to him. That is what accepting a cost looks like in every other regulated industry: you book it. You do not announce a Force, name three officials, leave the budget blank and call the result oversight.
Malaysia may yet produce a statute that is vague, underfunded and quietly shelved, and I will happily say so when it does. This week it remains the only government in the story that said the word "statute" out loud, in a room with a Hansard. Altman, for his part, has said the quiet part at a volume that ought to end the argument. He is not asking the world to accept some bad things. He is asking the world to accept them on his behalf, and to keep calling that agency.