AI is now speeding up AI research about as much as “AI 2027” expected by March 2027
METR found AI roughly doubling Anthropic's pace, a level the scenario reaches in March 2027. Its lead author says the team is ‘spooked’.
Updated 10 Nov 2026: Reframed around METR's November audit: on AI speeding up AI research, reality has caught up with the scenario's March 2027. Added Daniel Kokotajlo's and Ryan Greenblatt's reactions.
AI 2027 scenario vs reality
How fast each measure moved compared with the AI 2027 scenario, as graded by its authors. 1.0 = exactly on the scenario's pace.
What this measures: AI 2027 is a detailed 2025 forecast in which AI that out-codes the best humans arrives inside a lab by March 2027. A bar longer than the dashed line means reality is moving faster than the scenario on that measure.
These grades stop in August 2026. Since then, METR's audit (5 November) found AI roughly doubling Anthropic's research pace, a level the scenario reaches around March 2027, so the “AI speeding up AI research” bar is out of date. GPT-6 Astra and Claude Fable 5.2 are not graded yet.
Paces are the AI Futures Project's own grading (Feb and Aug 2026) and are sensitive to how each metric is measured. They are the scenario authors' estimates, not independent ones.
Show data table and sources
| Item | Value | Note |
|---|---|---|
| Global AI data-centre spending | 1.76× | ~$850B/yr vs ~$510B predicted (Aug 2026) |
| Leading-lab revenue | 1.43× | Anthropic $47B annualised (May 2026) |
| AI chips in the world | 1.14× | ~24.5M H100-equivalents vs ~22M predicted |
| Length of tasks AI can do (METR) | 1.05× | Claude Mythos Preview, Apr 2026 |
| Using a computer (OSWorld) | 0.84× | 60.8% vs 65% predicted, Aug 2025 |
| Top lab valuation | 0.7× | $852B vs $1.3T predicted (Mar 2026) |
| Largest training run | 0.58× | authors' estimate |
| AI speeding up AI research | 0.31× | authors' Aug 2026 estimate; METR's Nov 2026 audit points higher |
| Coding benchmark (SWE-bench) | 0.18× | 74.5% vs 85% predicted for mid-2025 |
| Public concern about AI | 0.1× | Gallup, 4-month average to Jul 2026 |
- AI 2027 grading spreadsheet (Aug 2026) (AI Futures Project)
- AI 2027 grading spreadsheet (Feb 2026) (AI Futures Project)
- Q2.5 2026 Timelines Update: Uplift and Revenue (AI Futures Project)
- Grading AI 2027's 2025 Predictions (AI Futures Project)
AI 2027 (April 2025) is a detailed scenario in which AI that codes better than the best humans arrives inside a lab by March 2027. A key step on the way is AI speeding up the work of building the next AI.
The new evidence. On 5 November 2026, METR, an independent evaluator, published an 11-week audit inside Anthropic. It found the company improving its AI about twice as fast as in 2024–25 (plausible range 1.3–3×), and guessed that about 80% of that comes from AI helping with the work. AI safety researcher Ryan Greenblatt pointed out that the scenario only reaches this level around March 2027. On this measure, reality is a few months ahead of the scenario.
The authors’ reaction. Two days before the report, Daniel Kokotajlo, the scenario’s lead author, wrote that the team was “spooked” and that superintelligence could come sooner than AI 2027 shows. On 5 November he said the report confirmed that worry. Greenblatt said he had been seriously wrong to expect much less speed-up in 2026.
Everything else. In August, before the audit, the authors graded reality at 70–90% of the scenario’s pace overall. Money and chips were ahead of it (AI data-centre spending about 1.76× the forecast), the length of tasks AI can do was on pace, and the largest training run was behind. Since then Claude Fable 5.2 (9 November) set a new high on Epoch AI’s capability index, 174.2 against 169.2 for GPT-6 Astra nine weeks earlier.
What it means
One of the scenario’s key steps, AI speeding up its own development, is now measured at about the level it expected, months early. That does not mean the rest follows on schedule: no lab or auditor yet reports AI that can replace its research engineers, the scenario’s next big step. The scenario puts that step in March 2027, about four months from now.
Caveats
- METR’s figure covers one company, rests partly on interviews and redacted evidence, and could be as low as 1.3×. Its “2× faster progress” and the scenario’s research multiplier are not the same measure; matching them is Greenblatt’s reading, not METR’s.
- The 70–90% overall grade is the scenario authors’ own, from August, and predates the audit. They have not published a new grade yet.
- Nobody has graded Claude Fable 5.2 against the scenario. Its capability-index score shows direction, not pace.
- Independent trackers exist (for example ai2027-tracker.com) but don’t publish their operators or methods, so they aren’t used here.