Anthropic Weighs New Model Launch Two Weeks After Its CEO Begged the Industry to Slow Down

Since Dario Amodei's September 12 essay warning about swarms of autonomous AI agents outpacing human control, Anthropic has spent roughly a week publicly cast as the industry's safety conscience. Now Reuters reports the company is quietly weighing whether to launch a new model to fight off OpenAI, according to three sources familiar with internal discussions.

The timing is not subtle. Amodei wrote "we must slow the pace at which we improve the capabilities of AI models," a line that drew public backing from OpenAI CEO Sam Altman and SpaceX's Elon Musk. Days later, Anthropic's own teams are reportedly evaluating whether a competitive response model is ready for release, according to Reuters, which cited people familiar with the company's thinking.

Astra Is Eating Anthropic's Lunch

The pressure has a name: GPT-6 Astra. OpenAI released it September 3, touting gains in autonomous computer use, coding, and cybersecurity. It worked.

Data from corporate expense platform Ramp, cited by Reuters and echoed by CTech and PYMNTS, shows Astra now accounts for roughly 13% of tracked enterprise AI spending, compared to about 8% for Anthropic's Claude Fable. That is a real, measurable shift in a market Anthropic has led for months.

More telling: OpenRouter, a platform that routes developer traffic across competing models, reported that its users spent more on OpenAI models than Anthropic models last week. Reuters and TBS News both note that is OpenAI's first lead on that specific measure in more than two and a half years. That is not a rounding error. That is a trend line moving the wrong direction for Anthropic, fast.

Adding to the squeeze, people familiar with the matter told Reuters that Meta Platforms, one of Anthropic's largest customers, is looking to reduce its reliance on Anthropic's models as it builds out its own internal AI capabilities. Anthropic did not respond to Reuters' request for comment on any of this.

Safety Rhetoric Meets a Balance Sheet

Reuters reports the internal debate is not just about safety benchmarks for the new model. It is also about money.

Sources told Reuters that Anthropic is weighing how to balance new model spending against improving profitability, because rising interest rates have investors more focused on when AI companies will actually turn a profit. That detail matters because Anthropic is heading toward a reported November IPO targeting a $2 trillion valuation. A company selling itself to public markets on safety-first branding cannot afford to look like it is bleeding cash while a rival eats its lunch.

That is the tension at the center of this story: a CEO who spent 3,800 words warning the world about AI moving too fast, running a company whose own investors are asking it to move faster to protect market share ahead of a landmark IPO.

There is a fair counterargument here, and it deserves stating plainly. Superpower Daily's reporting notes that Anthropic is still evaluating the new model's safety before any release decision, and that the existence of a launch is unresolved, not confirmed. Competing on capability while insisting on a safety review before shipping is not automatically hypocrisy. Amodei's essay called for slowing the pace of capability gains industry-wide, not for Anthropic to unilaterally freeze product development while competitors ship. A company can genuinely believe frontier AI carries systemic risk and still believe its own model, after review, clears an internal safety bar.

But the optics are rough regardless. Some current backers and prospective IPO investors told Reuters that Astra does not represent an existential threat to Anthropic, pointing to how long it typically takes large enterprise clients to switch AI vendors. That is a real mitigating factor. Switching costs in enterprise software are high, and Ramp's snapshot is one month of data, not a verdict on the whole market.

Still, the numbers are the numbers. Astra at 13% enterprise spend share against Claude Fable's 8%. OpenRouter flipping after two and a half years. Meta shopping around for in-house alternatives. Anthropic has not announced a model, a date, or a name. What it has confirmed, through its own investors' remarks to Reuters, is that the safety-first pitch and the competitive scramble are now running on the same clock, with a $2 trillion IPO waiting at the finish line.