Anthropic's AI Development Accelerates Despite Calls for Caution 2026

Anthropic's AI development is accelerating at an unprecedented rate, with recent analysis showing the company's models are improving nearly twice as fast as they were a year ago. This comes despite CEO Dario Amodei's mid-September warning that the industry might be moving too fast on AI.

The independent study, conducted by non-profit METR (Model Evaluation and Threat Research), found that Anthropic's models are now improving at a pace that is roughly double what it was between March 2024 and October 2025. The evaluation period ran from August 12 to October 29, meaning it began before Amodei's cautionary message.

What METR Found: AI Development Speed Doubles

METR's analysis focused on how quickly Anthropic improves its models' capabilities on well-defined tasks. Using the AECI (Anthropic's version of the Epoch Capabilities Index), which condenses a model's performance across a suite of benchmarks into a single number, the researchers estimated that since October 2025, the rate of improvement has grown 1.9 times faster.

The study provides a margin of error, suggesting the acceleration could be anywhere from 1.3 to 3 times faster. This range still represents a significant leap from the previous period.

Why the Sudden Acceleration? The Role of AI in AI Development

The key driver behind this acceleration appears to be AI itself. Anthropic is increasingly using AI to develop AI, creating a feedback loop that speeds up progress. This recursive self-improvement is a hallmark of advanced AI research, where models assist in tasks like code generation, testing, and optimization.

By leveraging their own models to automate parts of the development process, Anthropic can iterate more quickly and efficiently. This not only accelerates the pace of improvement but also potentially leads to more sophisticated models at a faster rate.

The Contradiction: Calls for Caution vs. Rapid Progress

Amodei's September message raised eyebrows, as it seemed to advocate for slowing down AI development. Some observers speculated that the call for caution might be less about safety and more about managing expectations or regulatory pressures. Regardless of the intent, the timing is notable: METR's study shows that Anthropic's internal progress has only accelerated since then.

This contradiction highlights a broader tension in the AI industry. Companies often publicly advocate for responsible development while privately racing to stay ahead. For Anthropic, the pressure to compete with rivals like OpenAI and Google DeepMind may outweigh public calls for restraint.

Implications for the AI Industry and Safety

The acceleration of AI development has significant implications. On one hand, faster progress could lead to breakthroughs in medicine, climate science, and other fields. On the other, it raises concerns about safety and control, as models become more capable and potentially harder to align with human values.

METR's findings also suggest that the pace of AI advancement may be underestimated by outside observers. If Anthropic's internal metrics are any indication, the industry could be on the cusp of even more rapid change. The background to this story is set out in Anthropic's AI Auditor Plan: METR and the Trust Problem. A related thread runs through OpenAI and Musk Back Calls to Slow Reckless AI Development. Earlier coverage in OpenAI Slows AI Development Pace After Rogue Agent Hack shows how the argument developed.

• Doubling speed: Anthropic's models are improving nearly twice as fast as in the previous period.
• AI-driven development: The acceleration is largely due to AI being used to develop AI, creating a feedback loop.
• Timing irony: The study period began before CEO Dario Amodei's call for caution, yet progress sped up.
• Wide margin: METR estimates the acceleration could be between 1.3 and 3 times faster.
• Industry pressure: Competition may be driving rapid progress despite public statements about safety.

What to Watch Next

As Anthropic continues to push the boundaries, the industry will be watching closely. Will the company maintain this accelerated pace, or will it heed its own calls for caution? The answer may depend on competitive dynamics and regulatory developments. A closer look at the same subject appears in UK House Prices Rise 0.1% in July as Rate Caution Persists. Anyone following the wider picture can start with BBC Shrinking at Accelerating Rate: Amol Rajan's Warning. The same theme surfaces in AI Development Pause Urged by Bernie Sanders in New Letter, which takes a different angle.

For now, the data suggests that AI development is not slowing down—it's speeding up. And Anthropic, despite its public stance, is at the forefront of this acceleration. Context on how this played out before sits in AI Stocks Fall After Development Slowdown Call Worries Investors. A companion piece, OpenAI and Anthropic Want to Pause AI Development: Is It a Smoke Screen, approaches the same subject from another angle.