Follow us on Google News
Artificial intelligence company Anthropic has withheld substantial portions of an independent research report examining how its use of AI is accelerating the development of more capable AI systems. The assessment, published on November 5, 2026, found that Anthropic’s model capabilities may be improving at roughly twice the previous rate, with researchers attributing much of the acceleration to AI-assisted research and development.
The study was conducted by the Model Evaluation and Threat Research (METR), an independent non-profit research organisation given extensive access to Anthropic’s internal data, including model benchmarks, pre-training metrics and analyses of AI-driven research acceleration.
Anthropic acknowledged redacting much of the report, arguing that publishing detailed findings could reveal commercially sensitive information and help competing AI laboratories accelerate their own development.
FCRF’s Chief AI Officer Certification Returns on November 14 to Prepare Future AI Leaders
Independent Researchers Find AI Development Accelerating
METR researchers assessed how Anthropic’s use of AI tools was influencing the pace of its research and development.
The study found that the company’s model capabilities on well-specified tasks were likely improving at approximately twice the rate recorded between March 2024 and October 2025.
Researchers estimated that around 80% of the acceleration could be attributed to Anthropic’s extensive use of AI in research and development activities.
They expressed confidence that the faster progress was not primarily explained by increased computing resources used for model training.
The assessment also found limited evidence of significant bottlenecks or diminishing returns that might otherwise slow further advances in AI capabilities.
According to METR, these findings raise the possibility that improvements in AI systems could enable further acceleration in research, creating a cycle of increasingly rapid technological development.
However, the researchers noted that AI systems were not yet capable of making fully autonomous contributions to AI research, despite showing early signs of such capabilities.
Anthropic Defends Decision to Withhold Key Findings
Anthropic co-founder and chief architect Sam McCandlish acknowledged that the company had removed substantial information from the published assessment.
In a post on X, McCandlish said the redactions made it more difficult for others to independently examine METR’s findings.
However, he argued that disclosing precisely what was driving the acceleration could provide other AI developers with information that might help them advance their own systems more quickly.
Anthropic maintained that the decision was necessary because the findings involved commercially sensitive information and could increase the very risks the independent assessment was intended to examine.
METR researchers said Anthropic had redacted a majority of the evidence they considered most informative for evaluating the potential risks associated with accelerating AI development.
The decision raised questions about the extent to which independent safety assessments can provide meaningful public scrutiny when the organisations being evaluated retain control over the information released.
Redactions Draw Criticism Amid Calls to Slow AI Progress
The disclosure prompted criticism from AI safety researchers, particularly because Anthropic has previously advocated slowing the development of increasingly powerful AI systems.
In September, Anthropic CEO Dario Amodei called for measures to pace frontier AI development and proposed stronger safeguards, including giving independent evaluators access comparable to that available to employees.
Daniel Kokotajlo, executive director of the AI Futures Project, questioned Anthropic’s decision to withhold important research findings while continuing to accelerate its own AI development.
He also challenged the company’s reliance on commercial sensitivity as a justification for limiting disclosure, given its public warnings about the potentially severe risks associated with advanced AI.
The findings come amid wider concerns about the pace of AI development and whether existing safety measures can keep up with increasingly capable systems.
While METR’s assessment indicates that AI-assisted research may already be contributing significantly to faster model improvements, the extensive redactions limit public access to the evidence supporting some of its conclusions.
The report highlights the continuing tension between commercial competition, independent oversight and transparency in the development of advanced artificial intelligence.
About the author — Suvedita Nath is a science student with a growing interest in cybercrime and digital safety. She writes on online activity, cyber threats, and technology-driven risks. Her work focuses on clarity, accuracy, and public awareness.
