METR's 1.9x estimate makes AI-accelerated AI development a measurable quantity

An independent evaluation credits AI-assisted research for most of Anthropic's recent acceleration, and planners should stop treating that possibility as a hypothesis 1.

METR has published an independent assessment built on a high-access evaluation of Anthropic that ran from 12 August to 29 October 2026 1. It estimates a 1.9x acceleration in Anthropic's benchmark progress, and that figure includes the company's recent internal-only models 1. Outside measurement of this kind is a meaningful step for a field that has relied largely on lab announcements, even though the precision of the estimate is limited 1.

The central judgement concerns cause as much as size. METR considers most of the current acceleration likely due to Anthropic's extensive use of AI in its own research and development 1. The same assessment acknowledges significant uncertainty and says other causes likely played a large role in late 2025, so the trend cannot be read as a steady climb from a single source 1. Laurenceo's reading is that this combination is the important finding: AI is plausibly speeding its own improvement now, yet the rate is volatile enough that extrapolating a single year would mislead.

Frontier labs are already making decisions on the same premise. OpenAI rated GPT-6 Cyber as Critical, its highest cyber risk tier, and released it only to approved Daybreak Red customers authorised for offensive security work 2. Its researchers used the model to find 81 previously unknown high-severity Firefox vulnerabilities and to help develop working exploits for 948 Linux kernel vulnerabilities 2. A capability that shortens the path from discovery to exploitation, arriving on a faster development curve, is precisely the case where fixed planning assumptions are most likely to fail.

Decision makers should act on three points. Enterprises, investors and regulators should widen their capability scenarios and watch whether METR or other evaluators repeat the exercise at other labs 1. Procurement and risk teams should avoid multi-year contracts that assume today's capability tiers will hold 1. Bermuda-based cyber underwriters should revisit exposure and accumulation assumptions as AI-driven exploit development becomes more capable 2.

“Outside measurement of how fast AI labs are speeding up their own work matters for every board that relies on these tools. The honest answer is that the number is uncertain, so plan for a range of outcomes rather than a single forecast.”

The Story

This editorial responds to the lead story of the briefing for 7 November 2026: METR estimates Anthropic's benchmark progress has accelerated 1.9x, driven mostly by AI-assisted R&D. Numbered links open the sourced story behind each point.

More Analysis

• Forecast Move · 7 November 2026Laurenceo raises odds that METR finds 2x AI acceleration by September 2027 to 50%
• The Laurenceo View · 6 November 2026Political licence, not chips, now sets the pace of AI infrastructure
• Forecast Move · 6 November 2026Laurenceo lowers Anthropic US listing forecast to 84% after IPO timing slips
• The Laurenceo View · 5 November 2026Frontier AI is now a capital problem as much as a technology one
• In Numbers · 31 October 2026AI and Bermuda in October 2026: security and data centre finance set the agenda
• Laurenceo Analysis · 1 November 2026Bermuda's AI risk sits on its balance sheets, not in its IT budgets

All analysis and editorial →

About The Laurenceo View

The Laurenceo View is the daily editorial of Laurenceo Intelligence: the publication's position on the day's lead story, argued only from Laurenceo's own sourced reporting. It is written by Laurenceo's research system under standing editorial direction set by Jorai Laurenceo, Founder & Executive Director, who can amend or withdraw it. How this is made.

Laurenceo Intelligence is the daily AI intelligence briefing published in Bermuda by Laurenceo, founded by Jorai Laurenceo. Every story is researched, source-verified and published each morning; how it works. General information only, not legal, regulatory or financial advice.