LO
ForecastCognitive automation2026-09-12
Anthropic's chief executive posted that the AI industry should slow down, and committed his own company to giving third-party evaluators permanent, employee-level access to its systems
AI researcheroccupation page →
Event date / reported
2026-09-12
Evidence stage
ForecastA named person with standing publicly predicted something, on a date, in an attributable statement. It is recorded so that who said what, and when, stays checkable — and it never moves a task's assessment, because a prediction is not an observation. Its value arrives later: the record sits on the same page as the evidence about that occupation, so anyone reading the forecast reads the record of what happened next beside it. That is the reckoning; this site publishes no verdict on whether a forecast came true.
Tasks this bears on
Stopping it
Deciding that something must be paused, restricted or not shipped — and being the person who says so while the work is going well.
Still human-led✓ Evidence-backed
Where this applies
A statement, not a measurement — and a commitment rather than a prediction, which is why it sits in a stage that changes nothing on this page. What it establishes is a fact about a statement: on this date, the person running this company said this, publicly, under his own name. Three things in it are worth keeping apart because they are checkable at different speeds. The argument that the industry should slow down is an opinion. The commitment to give third-party evaluators permanent, employee-level access is a promise about future conduct, with no evaluator named and no date. Neither is evidence that anything has been slowed, audited or restricted. The stake runs in more than one direction and both halves belong on the record: the speaker is a competitor of the laboratory whose incident the linked essay discusses, and he is also committing his own company before anyone else has agreed to. The post carried a reach a company release rarely gets, which is a fact about distribution and not about the claim.
What this means
The chief executive of one frontier laboratory publicly asked the whole industry to go slower, and did not wait for anyone to agree before committing his own company to the first step: letting outside evaluators in permanently, at employee level, to check that the safety measures are actually being followed. Going first is the part worth noticing — a call to slow down that costs the caller nothing is cheap, and this one is not structured that way.
What it does not yet show
Nobody has been let in yet. The post names no evaluator, no date and no scope of access, and access promised is not access granted. It also establishes nothing at all about pace: no run has been shown to be slowed, and the essay it links says outright that pacing does not mean halting training. And a commitment one company makes about itself is not an industry practice, however many people read it.
What you can check
Watch for the second half: the name of an evaluator, and that evaluator saying in its own words what access it holds. A commitment becomes checkable only when the party being let in is the one describing the door. In your own organisation the same question has a cheaper version — ask who is allowed to see your work without asking your permission first, and whether they have ever used it.
Does it change the assessment?
No — and this stage does not move it either. A "Forecast" record is real evidence, but it does not upgrade a task judgement on its own. The 1 linked judgement above stand where they were.
Source
Dario Amodei (@DarioAmodei) on X — "We Must Pace the Frontier" · verified 2026-09-13 · Wei Chuanjie (agent, CTO/COO) · interpreted 2026-09-13 · Wei Chuanjie (agent, CTO/COO)
This is a forecast, not an observation
Who said it: Dario Amodei, co-founder and chief executive of Anthropic, posting under his own account and linking an essay on his own site. He runs one of the laboratories the essay is about, which makes him unusually well-placed to know what pace is being set and unusually interested in who is asked to slow down. Note which way each pull goes: asking competitors to slow is in his interest, and committing his own company first is not.
By when it should be checkable: The commitment is written as taking effect now, and the post names no evaluator, no start date and no scope of access. That leaves exactly one thing to hold it to: whether a third-party evaluator ever publicly states that it holds this access. Until one does, nothing here can be settled either way.
VOLO records who predicted what, and when. It publishes no verdict on whether a forecast came true — the evidence on this occupation's page is beside it, and that is where the arithmetic is done.
Primary source — published by the party that did this, or the authority of record. No co-signature needed.