Several AI forecasting groups have moved their timelines for superhuman AI closer, citing new evidence that AI-assisted research may be accelerating progress. The underlying METR finding is striking, but much of the evidence behind it is not public, leaving an important gap between a worrying signal and a conclusion everyone can check.
METR Watch analysis
What happened
Defense One reports that Sentinel now expects superhuman AI around 2028 rather than 2031. The AI Futures Project, which previously outlined a scenario with superintelligence arriving in 2027, says progress is running ahead of that scenario’s assumptions.
Both groups changed their forecasts after a METR report on 5 November. METR found that the rate of improvement in AI models at Anthropic had roughly doubled since late last year, attributing much of the increase to AI agents helping with research. The report drew on Anthropic’s internal data, but Anthropic was allowed to redact information. METR said a majority of the evidence it considered most informative was removed.
Why it matters
If AI tools help researchers improve AI systems faster, they could speed up the work that produces the next generation of systems. That prospect is part of why the revised forecasts have attracted attention. But a faster measured pace of improvement is not, by itself, proof of a runaway self-improvement loop or a firm arrival date for superintelligence.
The missing evidence matters. It makes METR’s result harder for outside researchers to scrutinise, and it means the forecasts built on it carry uncertainty that a neat year in a headline can’t tidy away.
Our read
This is a meaningful change in expert expectations, not a countdown clock. The practical takeaway is that AI-assisted research deserves closer measurement: what tasks are being automated, how much faster they become, and whether the gains keep compounding. A forecast moving is a reason to examine the evidence, not to treat the most dramatic scenario as settled.
What to watch
- Whether METR publishes more detail that lets outside researchers assess its findings.
- Whether later measurements confirm that AI agents are accelerating AI research, and by how much.
- Whether Sentinel and the AI Futures Project revise their timelines again as new evidence arrives.
Discussion spark: Should AI forecasts that rely on partly redacted company data influence policy now, or should policymakers wait for results outsiders can scrutinise?
Sources and evidence
- Source update (19 November 2026, 21:22 UTC)
Independent WittyWires tracker for public updates about METR. Not affiliated with or endorsed by METR; this is not an official account.