An 11-week investigation by the AI safety group METR, published on Thursday, indicates that Anthropic’s models are improving at roughly double the pace seen a year prior.reported The report suggests that artificial intelligence systems are now the primary driver of this accelerated development cycle within the company.A new report says Anthropic is making its AI models much faster than before by using AI to help build them.METR’s 11-week audit indicates AI-assisted engineering is doubling Anthropic’s historical model improvement rates, independent of compute scaling.
Evidence: reported
Why a new report on Anthropic automating AI development is alarming researchersaifleet-verge-all
The figures
METR researchers stated that Anthropic has achieved roughly two years of historical progress over the last year. The organization noted that model improvement rates are approximately twice as quick as they were twelve months ago. These benchmark scores reflect the accelerated timeline observed during the investigation period.
Behind the figures
The findings stem from an 11-week investigation inside Anthropic, conducted as part of a broader evaluation of its research and development processes.reported METR researchers expressed confidence that the acceleration was not merely due to scaling up computing power, which has historically driven AI progress. Instead, they cited informal Slack polls and employee interviews to support the conclusion that AI tools are generating sizable productivity gains.
Evidence: reported
Why a new report on Anthropic automating AI development is alarming researchersaifleet-verge-all
What it means for your stack
The report highlights a shift where AI systems are building more powerful AI systems. This self-improving loop suggests that future model iterations may rely less on human-in-the-loop development and more on automated engineering processes. Developers should note that this acceleration comes with limited transparency, as Anthropic exercised its right to redact portions of the final report.
Building on this
Inference engineers can observe that the documented capabilities now include AI-assisted development workflows that significantly shorten iteration cycles. The interfaces and limits described imply a move toward autonomous model improvement, where the primary constraint is no longer just compute availability but the effectiveness of AI-driven optimization. Developers building on these models must account for the fact that the underlying training processes are becoming increasingly opaque due to corporate redactions, limiting external verification of specific architectural changes or safety guardrails implemented during this accelerated phase.
Why it matters
- Developers must adapt to faster release cycles driven by AI-assisted engineering, requiring more flexible integration strategies.
- The reliance on internal, redacted data limits external validation of model improvements, increasing uncertainty in third-party evaluations.
- The shift from compute-bound to AI-bound progress changes how engineers should anticipate future capability jumps in frontier models.
- Understanding that AI is now the primary driver of improvement helps teams prioritize automation in their own development pipelines.
Topics: Anthropic, METR, AI development, model training, safety research
Updated
Then / now / next
AI progress was historically driven primarily by scaling up computing power for training models.
METR reports that AI tools themselves are driving roughly twice the historical improvement pace.
Researchers will monitor whether this AI-driven acceleration continues without slowing down.
Story timeline
- Then AI progress was historically driven primarily by scaling up computing power for training models.
- 07 Nov 2026 Why a new report on Anthropic automating AI development is alarming researchers
- Now METR reports that AI tools themselves are driving roughly twice the historical improvement pace.
- Next Researchers will monitor whether this AI-driven acceleration continues without slowing down.
The strongest doubt
Counterpoint from the sources
Anthropic redacted portions of the report, raising questions about how much the public truly knows about the pace and risks of this development.
For readers
Developers
- AI tools are now the primary driver of model improvement at Anthropic.
- Expect faster iteration cycles as AI automates more of the development process.
Researchers
- The 11-week probe provides rare internal data on AI-driven R&D acceleration.
- Redactions in the report limit external verification of specific technical claims.
Investors
- Anthropic is preparing to go public amid reports of accelerated model progress.
- CEO Dario Amodei has called for slowing AI development despite internal speedups.
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.Questions readers ask
Is Anthropic’s acceleration due to more computing power?
METR researchers stated they are confident the acceleration is not simply due to scaling up computing power, but rather due to AI.
How long was the investigation?
The METR report followed an 11-week investigation inside Anthropic.
Can the public see all the details?
No, Anthropic was permitted to redact portions of the final report and appears to have exercised that power liberally.
Connected stories
Reading receipt
- METR found Anthropic’s models improved roughly twice as fast as a year ago.
- Researchers attribute most of this acceleration to AI rather than compute scaling.
- Anthropic redacted parts of the report, limiting public visibility into details.
Still open: How much critical information was lost due to Anthropic’s redactions?
Still open: Will this AI-driven acceleration continue indefinitely or plateau?