The team that wrote AI 2027, one of the more accurate forecasting documents of the past two years, is back with a positive vision for how humanity handles superintelligence. It’s called Plan A, and according to Don’t Worry About the Vase, which published a lengthy breakdown of the proposal and the reactions to it, the plan is already splitting the AI world into two camps that can barely talk to each other.
What stands out here isn’t the plan itself. It’s the credibility of the people behind it and the size of the gap between them and their critics.
Who’s making the prediction
Credentials matter when someone tells you to reshape global AI policy. Daniel Kokotajlo authored “What 2026 Looks Like,” which Don’t Worry About the Vase notes turned out “remarkably similar to what 2026 looks like.” His co-author Ryan Greenblatt, chief scientist at Redwood Research, ranked as the #2 most accurate AI forecaster in 2025 out of 413 entries. These aren’t random doomers. They have receipts.
That said, the author is careful, and so am I: past accuracy guarantees nothing. Don’t Worry About the Vase explicitly does not endorse Plan A. The point is that it deserves serious engagement, not a reflexive dismissal.
What Plan A actually says
Most people will never read the full document. They’ll compress it into a few sentences. Don’t Worry About the Vase predicts the five points that survive that compression, in order of attention:
- Slow down AI development.
- To do that, make a deal with China.
- Monitor the world’s major sources of compute.
- Use “mutually assured compute destruction” (a version of MAIM).
- Expect things to still move fast, with superintelligence by roughly 2040 and huge economic growth before then.
Axios boiled it down further: “To avoid dangerous outcomes from superintelligent AI, slow everything down.” Not wrong, but flat enough to invite misreading.
The real divide
The sharpest objection is also the simplest: superintelligence isn’t coming soon, so why pay steep costs to prevent it? Journalist Timothy B. Lee captured the standoff better than anyone. “There’s an epistemic chasm,” he wrote, “between those who think superintelligence implies near-omnipotence and those (like me) who don’t.” Each side, he added, finds the other’s intuition so alien it can’t quite believe the other is serious.
That’s the crux. Plan A only makes sense if you believe advanced AI is close and world-altering. Don’t Worry About the Vase agrees the labs believe it too. But if you don’t buy the premise, the honest response is to say Plan A is a bad idea and skip it entirely. The author calls that “a respectable response.”
Why it matters now
This is a preview of the policy fight coming to Washington and Beijing. The proposals in Plan A, compute monitoring, international agreements, deliberate slowdowns, are exactly the mechanisms governments will debate as AI capability climbs. The objections are already forming: America would never do it, China would never do it, this is a race and slowing down means losing, any talk of a “pause” is naive.
Kokotajlo’s answer to the futility charge is worth quoting: “We think it’s still good to recommend what would actually be good, even if you think that your audience is probably not going to listen.”
Takeaways for practitioners
- Know where you stand on timelines. Your position on AI regulation flows almost entirely from whether you think superintelligence is near. Get honest about that first.
- Watch the compute-governance debate. Monitoring chips and data centers is the lever both hawks and safety advocates keep reaching for. It will shape hardware access and compliance costs.
- Don’t confuse compression with the argument. The Axios one-liner is not the plan. If you’re going to argue for or against it, read past the headline.
- Track this crew’s forecasts. When people with this hit rate lay out a scenario, it’s worth understanding even if you bet against it.
Plan A isn’t a policy anyone is enacting tomorrow. It’s a marker for the argument that’s about to get much louder. Readers who want the full engagement, objections and all, can find it at the original source.