Key points
- AI Futures Project predicts a superhuman AI researcher will emerge between May and November 2027.
- METR audit found Anthropic doubling its model improvement speed since October 2025.
- Approximately 80 percent of Anthropic's speed boost stems from engineers using unreleased internal models.
What was released or claimed
The AI Futures Project published a document titled Q3.5 Timelines Update on November 10, 2026, stating that an intelligence explosion is now expected in 2027. The six authors unanimously agree that a Superhuman AI Researcher will likely appear between May and November of that year.A new report says AI will get smart enough to do science better than humans by late 2027.The AI Futures Project updates its forecast to a 2027 emergence of superhuman AI research capabilities, driven by observed acceleration at Anthropic. This entity would be so proficient in coding and scientific research that companies might prefer it over teams of human scientists.
Who it affects
The forecast directly impacts AI safety researchers and platform architects who must plan for systems capable of autonomous self-improvement. Tech companies face potential shifts in hiring strategies if automated AI becomes superior to human scientific teams. The rapid acceleration has intensified concerns among experts regarding the timing and safety of these developments.
What is still unknown
It remains unclear how exactly these superhuman capabilities will manifest in production environments or what specific safeguards will be in place by 2027. The precise technical mechanisms enabling the 1.7x speed-up in overall AI research progress are not fully detailed in the summary. Uncertainty persists regarding the broader societal impact of white-collar job displacement fears mentioned in passing.
What changes, what does not
The timeline for the intelligence explosion has shifted five months earlier than the original AI 2027 scenario predicted. The probability of seeing this acceleration by November 2026 was previously less than four percent in older models, but is now considered likely. Anthropic confirmed the rapid growth data reported by METR on the same day the audit was published.reported
Methods and limits
METR spent 11 weeks examining Anthropic’s internal model scores, training data, and coding activity. The audit revealed that Anthropic improved its models at roughly twice the speed compared to the period between March 2024 and October 2025. METR estimates that 80 percent of this boost comes from engineers using unreleased AI models for coding and data analysis. The AI Futures Project used these numbers to calculate the current acceleration rate.
Why it matters
- Platform architects must account for AI systems that can autonomously generate code and train newer models, changing development workflows.
- The 1.7x speed-up in AI research progress suggests that current planning horizons for model capability may be too conservative.
- The reliance on unreleased internal models for development indicates a new feedback loop that could accelerate capability gains unpredictably.
- Safety concerns are intensifying as the timeline for superhuman AI research capabilities moves closer to the present.
Topics: AI models, Anthropic, METR, AI safety, research timelines
Updated
Then / now / next
The intelligence explosion was previously thought to be years away with low probability of near-term acceleration.
Researchers expect it in 2027, citing doubled model progress at Anthropic.
Watch for further updates on how widespread the use of internal AI for development becomes across the industry.
Story timeline
- Then The intelligence explosion was previously thought to be years away with low probability of near-term acceleration.
- Now Researchers expect it in 2027, citing doubled model progress at Anthropic.
- Next Watch for further updates on how widespread the use of internal AI for development becomes across the industry.
The strongest doubt
Counterpoint from the sources
Lead author Daniel Kokotajlo expressed fear, sadness, and anger at the rapid acceleration, noting it happens five months ahead of schedule.
For readers
Researchers
- Review assumptions about AI self-improvement timelines in light of the 2027 forecast.
- Analyze the impact of internal AI usage on development velocity.
Policy
- Consider regulatory frameworks for systems that can autonomously conduct scientific research.
- Monitor safety concerns as the timeline for superhuman AI compresses.
Developers
- Evaluate how using internal AI models for coding might accelerate or complicate workflows.
- Prepare for potential shifts in the value of human scientific teams.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
Intelligence explosion
A scenario where AI becomes smart enough to build even smarter AI autonomously.
Superhuman AI Researcher
An AI system superior to human teams in coding and scientific research.
METR
Model Evaluation and Threat Research, the group that audited Anthropic.
Questions readers ask
When do researchers expect a superhuman AI researcher to emerge?
The AI Futures Project forecasts emergence between May and November 2027.
What caused the acceleration in AI progress?
METR found Anthropic doubled its improvement speed, largely due to engineers using unreleased internal AI models.
Did Anthropic confirm the audit findings?
Yes, Anthropic confirmed the rapid growth data on the same day METR published its report.
Connected stories
Reading receipt
- AI Futures Project predicts a superhuman AI researcher will emerge between May and November 2027.
- METR audit found Anthropic doubling its model improvement speed since October 2025.
- Approximately 80 percent of Anthropic's speed boost stems from engineers using unreleased internal models.
Still open: How will the 1.7x speed-up in AI research progress impact other companies beyond Anthropic?
Still open: What specific safety measures are being developed to address the earlier timeline?