5 Questions About METR’s Findings On AI-Assisted Development At Anthropic
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

METR’s November 5 assessment estimates Anthropic’s model capability progress accelerated about 1.9 times after October 2025, with a 95% confidence interval of 1.5x to 2.8x. METR attributes perhaps 80% of that acceleration to AI assistance, but says Anthropic redacted much of its most informative evidence, limiting public scrutiny.

METR, a nonprofit that evaluates frontier AI systems, said on November 5 that Anthropic’s model capability progress has accelerated to about 1.9 times its earlier pace, based on an 11-week assessment of internal company data. METR estimates that AI use likely accounts for most of the increase, but says Anthropic redacted a majority of the evidence it found most informative, leaving the public with limited ability to check the analysis.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

METR compared Anthropic’s progress before and after an estimated October 7, 2025, turning point. Its measure, the Anthropic version of the Epoch Capabilities Index, rose at about 15 points per year before that date and about 29 points per year afterward. METR’s acceleration estimate is 1.9x, with a 95% confidence interval of 1.5x to 2.8x. The organization said it would not be surprised by an estimate between 1.3x and 3x.

The assessment says AI may account for about 80% of the acceleration. METR said it was confident the change did not result from faster growth in training compute. Anthropic’s internal measurement, as described in the report, found Claude led 26% of its AI research and development work in August 2026, compared with less than 1% in February. The company’s main internal platform reportedly had about 30,000 agents working at a time.

METR’s estimate is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card. METR said the inclusion of newer, unreleased models was the main reason for the difference. Looking across frontier models from all developers, rather than Anthropic alone, METR estimated a smaller acceleration: from about 17 to 24 index points per year, or 1.4x.

At a glance
reportWhen: Published November 5, 2026; assessment…
The developmentMETR published an assessment based on 11 weeks of access to Anthropic’s internal AI research and development data, estimating a roughly 1.9x acceleration in model progress.
AI research · Briefing

METR’s Anthropic findings in five questions

METR estimates that Anthropic’s model capability progress accelerated after October 2025, with AI assistance accounting for much of the increase. Redactions limit how much of the analysis the public can verify.

1.9×
Estimated acceleration
1.5×–2.8×
95% confidence interval
≈80%
Acceleration METR attributes to AI
26%
Claude-led R&D work in August 2026

What did METR measure?

Its Anthropic capability index rose from about 15 points per year to about 29 after an estimated October 7, 2025 turning point.

Before
  • About 15 index points per year
  • Earlier pace of model capability progress
After
  • About 29 index points per year
  • Estimated acceleration of 1.9×

Does this mean AI builds AI alone?

No. The report concerns AI assistance with research and engineering; it does not establish fully autonomous development of successor models.

AI assistance

R&D support

METR estimates AI use may account for most of the acceleration, but this is not a directly observed share of every research decision.

Autonomy

Not established

Anthropic says it has not reached a system that fully autonomously designs and develops its successor.

Why is public verification limited?

METR had internal access, but much of the evidence it found most informative was withheld from public view.

  • Anthropic redacted a majority of that evidence.
  • METR dropped a planned check of how compute is divided between safety and capability work.
  • The estimate covers a selected capability index and may not describe all AI research or persist over time.

What happens next?

The focus is whether Anthropic concludes its Responsible Scaling Policy threshold has been met and what safeguards would follow.

1
Assess the threshold

METR judged a crossing plausible by July 15; Anthropic said it may imminently meet its definition.

2
Clarify safeguards

The available information does not confirm a decision date or specify a new public evaluation.

3
Enable scrutiny

Further disclosure could help researchers assess the index, AI’s role, and withheld evidence.

CaveatThe 1.9× figure measures capability progress on specified tasks; it does not mean researchers are twice as productive overall.
The takeaway
METR’s evidence points to faster Anthropic model progress with substantial AI assistance, while redactions leave key parts of the finding difficult to verify publicly.

How AI Use Could Change R&D Timelines

The report addresses whether AI tools are speeding up the research and engineering that produce more capable AI systems. If the acceleration estimate is broadly right, labs could reach capability milestones faster than they would through conventional scaling alone. That could shorten the time available to evaluate new systems and prepare safeguards.

The number measures progress on specified capability tasks; it does not mean researchers are twice as productive overall, or that AI systems are independently designing their successors. METR’s attribution that AI accounts for perhaps 80% is an estimate, not a directly observed share of every research decision. The findings are consequential, but their interpretation depends on the measure, the selected models and evidence the public cannot inspect.

Products Worth Considering

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

As an affiliate, we earn on qualifying purchases.

Anthropic’s Twofold Progress Threshold

Anthropic’s Responsible Scaling Policy, version 3.4, took effect on July 8, 2026. It defines an “automated R&D” trigger that includes a doubling of progress pace when that increase is plausibly substantially attributable to research or engineering automation. The policy treats the threshold as a potential early warning; meeting it would not by itself prove that fully autonomous AI research has arrived.

METR’s work differed from typical outside model evaluations because its researchers had access to internal information, including benchmark results for unreleased models, training metrics, staff surveys and about 15 interviews. The assessment involved roughly 2.5 full-time-equivalent staff from August 12 through October 29, working on Anthropic devices. METR says it received no payment. Anthropic supplied laptops and free tokens and retained the right to redact non-public information.

Anthropic said in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress. METR also judged it plausible that the company’s early-warning threshold had been crossed by July 15. These are related assessments, but neither statement establishes that AI can autonomously develop a successor model.

“A 2x acceleration means that 2 years of historical progress is made in 1 year.”

— METR

What the Redactions Leave Unverified

Anthropic redacted what METR described as a majority of the evidence it found most informative. That restricts independent public review of how the data support the acceleration estimate and the attribution to AI use. METR also dropped a planned check of how Anthropic divides compute between safety work and capability development.

The reported acceleration concerns a selected index of benchmark performance and a change point estimated by METR. The report does not establish that the same pace applies across all AI research, nor that the estimate will persist. The provided material does not detail the full reasoning behind the redactions or provide enough unredacted evidence to resolve those limits.

Anthropic’s Policy Threshold in Focus

The immediate next question is whether Anthropic concludes that its Responsible Scaling Policy threshold has been met and what safeguards it would apply if so. METR’s report says crossing the threshold was plausible by July 15, while Anthropic’s filing said the threshold may be met imminently. The available information does not specify a confirmed decision date or a new public evaluation.

Further disclosure could help outside researchers assess the index, the role of AI assistance and the withheld evidence. Until then, the 1.9x estimate remains an external assessment informed by internal access, with important parts of its evidentiary basis unavailable for public examination.

Key Questions

What did METR estimate?

METR estimated that Anthropic’s model capability progress accelerated by about 1.9x after an October 2025 turning point, with a 95% confidence interval of 1.5x to 2.8x.

Does the report say AI is building AI without human involvement?

No. The report concerns AI assistance with research and engineering. Anthropic says it has not reached a system that fully autonomously designs and develops its successor.

How much of the acceleration does METR attribute to AI?

METR estimates that perhaps 80% is due to AI use. The report presents this as an estimate, and much of the evidence METR considered most informative was redacted.

Why is the finding difficult to verify publicly?

Anthropic redacted a majority of the evidence METR found most informative. METR also dropped a planned check of how the company splits compute between safety and capabilities work.

Source: Thorsten Meyer AI

You May Also Like

What OpenAI Dev Day 2026 Means For The AI Ecosystem

OpenAI announced agents, plugin changes and new APIs at DevDay, positioning ChatGPT as a platform for software and automated work.

I Wasn’t Allowed Prompting ChatGPT During My Chalk Talk: This Is Discrimination (2025)

A teacher alleges discrimination after being barred from prompting ChatGPT during a classroom presentation, raising concerns about access and fairness.

Why My Favorite AI Test Result Isn’t Running My News Pipeline

A publisher’s test of 71 model configurations found GPT-6 Astra scored highest, but cheaper GPT-6 Luna is used in production.

How An AI Chatbot’s False Murder Tip Reached Philadelphia Police

An Ars Technica headline reports a false murder tip involving Claude and Philadelphia police. The available material leaves the submission process and response unknown.