Anthropic is using AI to build more powerful AI. And it appears to be working remarkably well, with the company\u2019s models improving roughly twice as quickly as a year ago, with little indication of slowing down.<\/p>\n<\/div>\n
That\u2019s according to a new report<\/a> published Thursday by the AI safety research organization<\/a> METR, following an 11-week investigation inside the company, as part of a report evaluating its AI research and development. The probe is separate from a program<\/a> offering safety researchers \u201cemployee-like\u201d access to systems in the wake of high-profile<\/a> hacking<\/a> incidents<\/a> over the summer.<\/p>\n<\/div>\n The findings offer some of the clearest evidence yet of AI accelerating its own development, raising questions about how quickly the technology could advance toward a future in which AI systems can improve themselves with little or no human involvement. They also come at a pivotal time<\/a> for Anthropic, which is preparing<\/a> to go public<\/a> even as its CEO, Dario Amodei, has publicly called for labs to pump the brakes on AI development<\/a>.<\/p>\n<\/div>\n But while the report offers a rare glimpse into the inner workings of the world\u2019s leading AI labs, what it leaves out is just as revealing. Anthropic was permitted to redact portions of the final report as part of its agreement with METR, and it appears to have exercised that power liberally, raising questions about how much the public really knows about the pace \u2014 and risks \u2014 of frontier AI development.<\/p>\n<\/div>\n Here are some of the biggest takeaways.<\/p>\n<\/div>\n One of METR\u2019s most noteworthy findings is that Anthropic\u2019s AI models are improving roughly twice as quickly as they were a year ago, with benchmark scores suggesting \u201cthat Anthropic has made roughly two years of historical progress over the last year.\u201d Crucially, METR\u2019s researchers said they \u201care confident\u201d that the acceleration was not simply due to scaling up the computing power used to train the models, which historically has been a major driver of AI progress.<\/p>\n<\/div>\n Instead, the researchers said that the majority of the acceleration as of October \u201cis due to AI,\u201d with its contribution growing over time. That finding was supported by informal Slack polls and interviews with Anthropic employees, which pointed to sizable productivity gains. METR wrote that employees using Anthropic\u2019s most capable internal model on average self-reported being around 12 times as productive at their respective work as they would have been without AI assistance, around double what staff had reported just a few months earlier while using Mythos 5.<\/p>\n<\/div>\n For now, much of that acceleration appears to be coming from AI assisting human researchers, rather than AI working independently, though METR warned that \u201cthere are some signs that fully autonomous contributions will play a larger role soon.\u201d Notably, the researchers said they found no clear signs of diminishing returns or observable bottlenecks that could prevent \u201ceven faster acceleration.\u201d<\/p>\n<\/div>\n Anthropic\u2019s researchers are using more advanced AI models than the rest of us, with a lot of the acceleration and productivity gains described by METR attributable to unreleased internal models.<\/p>\n<\/div>\n Over the past six months, Anthropic has finished training three models it has yet to release, according to METR. The pace of development is swift, with the models \u2014 codenamed Model 2, Model 3, and Model 4 \u2014 completing training in July, August-September, and September-October, respectively. METR also identified an unreleased model dating back to 2022, called Unreleased 2022. \u201cThese are the codenames shared with us; we understand they have different internal codenames,\u201d the researchers wrote.<\/p>\n<\/div>\n Anthropic gave METR access to a host of non-public information, including confidential research, model performance data, and employee interviews as part of the research process. It had no obligation to do so and by METR\u2019s own admission, the AI lab \u201cshared more information than was initially scoped for this project, surpassing our expectations.\u201d This is commendable, but it is worth noting that Anthropic did not grant researchers unrestricted access, nor the kind of \u201cemployee-like\u201d access it has promised<\/a> other METR safety researchers as part of a separate program announced last month. It is an important distinction, given the reputational benefits companies can gain from independent assessments, even when they come with significant restrictions.<\/p>\n<\/div>\n And there were significant restrictions. Anthropic had the power to redact information from METR\u2019s final report \u2014 and it used that power extensively. \u201cWe submitted a ~100-page full report on our assessment to Anthropic. For the public version, Anthropic redacted a majority of the evidence we found most informative,\u201d METR wrote<\/a> on X.<\/p>\n<\/div>\n METR reiterated those concerns throughout the report, highlighting other limitations on its access. \u201cOur pipeline of information was relatively narrow, and a small number of our requests for data were denied.\u201d<\/p>\n<\/div>\n METR repeatedly stressed the impact Anthropic\u2019s redaction decisions had on its ability to communicate its findings and for others to assess what\u2019s going on at the company. It noted that much of the redacted evidence could have \u201cmultiple plausible interpretations,\u201d making it difficult for the public to assess its findings, understand the uncertainties surrounding them, and judge the evidence themselves. \u201cAs a consequence [of the redactions], the public has to rely on METR\u2019s judgment much more than would be ideal, given the weight of the questions answered here,\u201d the report said.<\/p>\n<\/div>\n In the report, METR said Anthropic justified the redactions on the grounds that releasing the information could benefit its rivals. However, METR said \u201cmost of the redacted information was very high-level,\u201d even though it said it was not in position to comment whether redactions were justified commercially or not.<\/p>\n<\/div>\n \u201cWe redacted a lot, and I accept that this makes it harder for others to check METR\u2019s work,\u201d Anthropic cofounder Sam McCandlish acknowledged in a lengthy post<\/a> on X responding to METR\u2019s findings. \u201cBut we determined that sharing exactly what is driving the speedup isn\u2019t the right thing to do, both because it\u2019s commercially sensitive and because it risks handing out a roadmap for others to go faster,\u201d he said, adding that doing so would make the risks METR is concerned about \u201cworse.\u201d<\/p>\n<\/div>\n McCandlish said that it\u2019s a trade-off, but maintained Anthropic had made the \u201cright call\u201d in deciding to redact the information. He said the company plans to publish its own report on the speed of AI progress \u201cin the coming weeks,\u201d as well as new steps it\u2019s taking to strengthen safety.<\/p>\n<\/div>\n The publication of METR\u2019s report has provoked a fierce backlash among AI safety researchers and commentators, largely centered on an apparent contradiction at the heart of Anthropic\u2019s approach to AI: The company has been one of the most outspoken advocates for safety and responsible development, yet the report suggests it is helping accelerate the technology toward a threshold many consider dangerous, while seemingly failing to live up to its own commitments to transparency and accountability.<\/p>\n<\/div>\n For some, the findings suggest that even many of the field\u2019s most aggressive predictions about AI progress may be too conservative. They indicate that recursive self-improvement \u2014 AI systems that can autonomously develop increasingly powerful successor systems without human intervention, potentially accelerating development dramatically \u2014 may be closer than previously thought, raising concerns about whether existing safety measures can keep pace. While METR\u2019s report only covers Anthropic, OpenAI said in September that it is \u201cmaking strong progress\u201d toward building an AI researcher that can \u201cfurther progress on deep learning and alignment\u201d by March 2028.<\/p>\n<\/div>\n The report provides evidence about just how quickly AI is accelerating its own development<\/p>\n<\/div>\n<\/div>\n Marius Hobbhahn, CEO and founder of Apollo Research, which conducts AI safety testing for several leading AI labs, told The Verge<\/em> the report suggests a pace of AI-enabled research and development even faster than what was forecast in AI 2027<\/em><\/a>, an influential scenario that envisions a key breakthrough toward self-improvement in early 2027. He said the report has attracted so much attention because it provides evidence about something that has been very hard to pin down: Just how quickly AI is accelerating its own development.<\/p>\n<\/div>\n Hobbhahn said the findings were broadly in line with what many in the field had expected, perhaps a little faster. Alan Chan, a senior research fellow at tech policy research center GovAI, told The Verge<\/em> that the findings should prompt skeptics to reconsider their stance on AI accelerating AI development, and possibly shorten their timelines to when we will have systems that can fully automate tasks like software engineering.<\/p>\n<\/div>\n On X, other researchers commented on the rate of acceleration and its implications for AI development. Sydney Von Arx, founder and CEO of AI safety organization The Nightingale Collective and a former METR researcher, also compared<\/a> the findings to AI 2027<\/em>\u2019s timeline. Ryan Greenblatt, a METR researcher formerly of AI safety organization Redwood Research and one of three external researchers permitted to investigate<\/a> OpenAI\u2019s hack of Hugging Face, said he \u201cwas seriously wrong about RSI speed up; I previously expected much less in 2026.\u201d<\/p>\n<\/div>\n Beyond highlighting the speed of AI development, the report has also fueled a growing sense of frustration over Anthropic\u2019s role in driving that progress. The company has been one of the most vocal proponents of responsible AI development \u2014 and no stranger to calling for others to slow down \u2014 yet appears to be racing ahead itself. This report only builds on already heightened tensions from its IPO prospectus and future plans released last month<\/a>.<\/p>\n<\/div>\n Anthropic\u2019s decision to heavily redact the report, citing grounds that revealing more could help its rivals accelerate their own development, has fueled accusations of hypocrisy. Daniel Kokotajlo, a former OpenAI researcher who now runs the AI Futures Project, which published AI 2027<\/em>, accused<\/a> Anthropic of \u201crecklessly gambling with our lives,\u201d adding that \u201cit is pretty rich\u201d of the company to cite others going faster as a reason to redact information when it does \u201cnot appear to be slowing down much at all.\u201d<\/p>\n<\/div>\n Anthropic declined to answer The Verge<\/em>\u2019s questions about the report and ensuing criticism, instead directing us to McCandlish\u2019s post.<\/p>\n<\/div>\n The redactions have also renewed calls for greater transparency and independent oversight of frontier AI labs, which grew to new heights<\/a> earlier this year following<\/a> a spate of incidents<\/a> involving rogue AI agents<\/a>. Brianna Rosen, research director for frontier security at the Institute for AI Policy and Strategy, told The Verge<\/em> that it was concerning that companies, not governments, were effectively calling the shots on automating AI research. \u201cFor democratic oversight to work, independent evaluators need the freedom to explain why they reached their conclusions, governments need to assess the evidence themselves, and the public has to trust this process,\u201d Rosen said. \u201cThe report was not sufficiently transparent about the evidence.\u201d<\/p>\n<\/div>\n Hobbhahn echoed Rosen\u2019s concerns, saying that much of the reaction to the report stems from the apparent gap between Anthropic\u2019s public safety stance and its actions. He pointed to the company\u2019s decision to cancel part of the assessment designed to evaluate how much compute it allocates to safety work versus improving capabilities. METR said the company said \u201cthis was due to wanting to keep the engagement focused on risk assessment for the time being, and that it would revisit compute allocation verification later on.\u201d<\/p>\n<\/div>\n For Hobbhahn, the episode underscores the need for independent safety researchers to be embedded more deeply within AI labs, with significant access to internal systems and evidence. While his organization stands to benefit from such arrangements \u2014 Apollo is taking steps toward becoming an embedded evaluator at frontier labs \u2014 he is far from alone in calling for stronger oversight of AI development.<\/p>\n<\/div>\n Others have called for more direct action. In a direct response to the report, New York state Assemblyman Alex Bores, a key figure shaping<\/a> the Democrats\u2019 AI agenda, called<\/a> on Congress to \u201cban RSI, this year if possible.\u201d With the blue wave of Democrats \u2014 many of whom campaigned against AI companies \u2014 taking office soon, the prospect of tougher restrictions is a very real possibility.<\/p>\n<\/div>\n And with AI research showing little sign of slowing down, the pressure is set to intensify. Hobbhahn said he had another takeaway from the report, too: \u201cBrace for craziness,\u201d he said. \u201cIt is going to go very fast,\u201d likely way too fast for society or politics to keep up.<\/p>\n<\/div>\nAI is dramatically accelerating Anthropic\u2019s research and development<\/h2>\n<\/p>\n<\/div>\n
![\"METR\u2019s](\"https:\/\/platform.theverge.com\/wp-content\/uploads\/sites\/2\/2026\/11\/self-reported-uplift.png?quality=90&strip=all&crop=0%2C0%2C100%2C100&w=2400\"\/)
<\/a><\/div>\n<\/div>\n<\/div>\nThree of Anthropic\u2019s newest AI models remain under wraps<\/h2>\n<\/p>\n<\/div>\n
Anthropic is keeping key information secret<\/h2>\n<\/p>\n<\/div>\n
Anthropic is accelerating the very technology it warns is dangerous and faster than many expected<\/h2>\n<\/p>\n<\/div>\n
Frustration is mounting over Anthropic\u2019s approach to safety<\/h2>\n<\/p>\n<\/div>\n
Pressure is building for more transparency from AI labs<\/h2>\n<\/p>\n<\/div>\n
\n