Anthropic AI Automation Report Raises Safety Concerns
Left 67%
Center 0%
Right 33%
2 left · 0 center · 1 right
What happened
METR, an independent AI safety research group, released a report on Thursday after an 11-week investigation inside the AI company Anthropic's San Francisco headquarters, where six researchers interviewed employees, monitored internal Slack channels, and examined unreleased models on company hardware. The report found that Anthropic's models improved roughly twice as quickly as a year earlier — about two years of historical progress in the last year — and that researchers were confident the acceleration was not simply from scaling compute, with the majority of the speedup as of October due to AI. Over the prior six months Anthropic finished training three models not yet released to the public, completed in July, August–September, and September–October, and internal Slack-poll data showed employees using the most capable internal model self-reporting roughly 12 times the productivity they would have had without AI assistance. METR said it submitted an approximately 100-page assessment and that Anthropic redacted a majority of the evidence the researchers found most informative; Anthropic cofounder Sam McCandlish said sharing exactly what is driving the speedup was commercially sensitive and would risk handing others a roadmap to go faster. METR described the acceleration as accomplished primarily via AI assistance rather than fully autonomous contributions, while noting some signs that fully autonomous contributions would play a larger role soon and that it did not see observed bottlenecks that would prevent even faster acceleration.
Sourcing check · Fully on the record · 3 articles
2 of 2 articles cite a named source for the story. It was on the record from the first report.
What these terms mean and how they're counted
- On the record · names a source 2 of 2
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: METR
- Primary evidence · cites a document or data 1 of 2
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: report
- Credits another outlet 0 of 2
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 0 of 2
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Nov 5, 20:40 UTC by Business Insider, naming its source. First document the next day, from The Verge (report).
- Not counted 1
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
Omitted — what each side leaves out
Unpacked
Right-leaning coverage never mentions the METR audit that left-leaning coverage treats as the news. Independent researchers inside Anthropic reported roughly two years of historical progress, driven mostly by AI assistance, and said the company then redacted most of the evidence they found most informative, according to The Verge and Business Insider. Right-leaning coverage instead offers a general warning about recursive self-improvement and a training pause, with no audit and nothing about unreleased models. A reader of only that side would not know an outside team had been inside the lab.
Left-leaning coverage leads with the audit and the redactions. Right-leaning coverage hangs the speedup story on scripture, with the Daily Wire warning against a new "Tower of Babel."
Would the redacted pages still support the claim that AI, not extra compute, is now the main cause of Anthropic's speedup?
Bottom line
A Daily Wire reader never hears METR's name, so the redaction fight The Verge put at the center of the Anthropic story does not exist for them.
The receipts — 5 facts one side's coverage lacks
- METR found Anthropic had made roughly two years of historical progress over the last year
- METR said Anthropic redacted a majority of the evidence the researchers found most informative
- Anthropic finished training three models in the past six months that it has not released
- Staff using the most capable internal model self-reported being about 12 times as productive
- Anthropic paused training its newest model for over a month because of safety concerns
The Left View
Left-leaning outlets treated the report as alarming evidence that recursive self-improvement — AI systems increasingly able to improve successor systems with less human involvement — is arriving faster than many expected, with safety researchers comparing the pace to the AI 2027 scenario, which envisions a key self-improvement breakthrough in early 2027, and finding Anthropic possibly ahead of it. Business Insider and The Verge stressed a contradiction between Anthropic's public calls for labs to slow development and its internal use of unreleased models to speed research, quoting former employee Jacob Coxon that the leading labs are "racing straight to self-improving superintelligence and gambling with our lives." They presented the redactions as blocking independent scrutiny: METR wrote that "the public has to rely on METR's judgment much more than would be ideal," and Daniel Kokotajlo of the AI Futures Project called it "pretty rich" to withhold details lest rivals go faster when Anthropic "does not appear to be slowing down much at all." Coverage amplified demands for outside control of the process, including Brianna Rosen's warning that companies, not governments, were "effectively calling the shots on automating AI research," and New York Assemblyman Alex Bores's call for Congress to "ban RSI, this year if possible."
The Right View
The Daily Wire cast the same underlying shift — AI taking a larger role in building the next generation of AI — as a moral and spiritual crisis of human limits, arguing that labs are not merely building a new Tower of Babel but "outsourcing the construction to AI." Campus pastor Michael Grayston, citing Pope Leo XIV's May encyclical warning against a new Babel, wrote that Anthropic's own caution that models accelerating their own development "could make it more challenging for humans to understand or control these systems" shows why "human developers [must] remain in control," and treated a reported month-plus pause in training Anthropic's newest model over safety concerns as a reminder that such control is not guaranteed. The piece argued the race for superintelligence "began without the guardrails needed," that self-regulation by "powerful, well-resourced companies" is not enough, and that statements such as OpenAI CEO Sam Altman's "Your children will never be smarter than AI" reflect a "lack of responsibility"; it closed by asking whether people will "let AI builders move beyond their control before putting sensible guardrails in place."
Our Take (balanced)
The left's strongest argument is that METR's access produced measurable signs of AI-driven acceleration — two years of progress in one, large self-reported productivity gains, a stack of unreleased models, and no clear bottlenecks — while heavy redactions and Anthropic's concurrent public safety messaging leave outsiders unable to test whether risk controls match the pace, a problem METR stated directly when it said the public must rely on its judgment "much more than would be ideal." The right's strongest argument is that the same automation trend is a control problem, not only a disclosure problem: Anthropic has already warned that AI-sped development could make systems harder for humans to understand or control, so the danger is handing construction itself to systems that "will not rest, will not stop." The central unresolved tension is whether the story is primarily a breakdown of transparency and self-restraint at a safety-branded lab whose actions undercut its warnings, or a broader race in which capability is being delegated to AI faster than any shared authority — corporate, governmental, or moral — can still set a limit.
3 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.