In the months before a closely watched public offering, Anthropic finds itself at a crossroads that few companies are forced to navigate so openly: its chief executive has just warned the world that frontier AI development moves too fast, and yet the market is moving faster still. OpenAI's latest model has begun pulling enterprise spending away from Anthropic's own flagship system, and the company is now quietly debating whether to release a new model of its own. The decision, still unresolved, will reveal whether safety advocacy and competitive survival can coexist — or whether, when the stak
Anthropic weighs new AI model release despite CEO's calls for development slowdown
Caught between two competing imperatives: the market and the public stance
Mark
So Anthropic's CEO just published this long essay about slowing down AI development, and now the company is thinking about releasing a new model anyway. That seems contradictory.
Mimi
It does on the surface, but the context matters. OpenAI's GPT-6 Astra came out in early September and immediately grabbed market share—13 percent of enterprise spending versus Anthropic's 8 percent. That's a real shift. Amodei's essay came out two weeks after that launch, so he was already responding to competitive pressure when he wrote about systemic risks.
Luke
But here's what I want to know: did Amodei write the essay knowing Anthropic was considering a response launch? Or did the internal discussions start after the essay? The reporting doesn't say. That's a material difference.
Mimi
Fair point. The reporting is based on three unnamed sources discussing internal deliberations, so we don't have a clear timeline of when the decision-making actually began. What we do know is that teams are currently evaluating safety benchmarks while deciding whether to greenlight a launch.
Mark
What does Anthropic's financial position actually tell us here? Are they under real pressure, or are they just trying to stay ahead?
Mimi
Their annualized revenue run rate hit $65 billion by late July—that's a massive jump from $9 billion at the end of 2025. OpenAI is at $40 billion. So Anthropic is still ahead on revenue, which suggests they're not desperate.
Luke
But revenue run rate and actual revenue are different things. And the reporting doesn't explain how those numbers were calculated or whether they're sustainable. Also, investors are saying Astra doesn't pose an existential threat, but that's what investors say when they want to keep both companies' valuations up before IPOs. That's not independent analysis.
Mimi
True. The reporting does note that Meta, one of Anthropic's top customers, is scaling back its reliance on Claude and building its own internal models. That's a real structural challenge that goes beyond just competing with OpenAI.
Mark
So the question isn't really whether Anthropic should release a model—it's whether Amodei's safety argument was genuine or just positioning?
Luke
Or both. People can believe something is risky and still feel compelled to do it because the market demands it. The reporting doesn't resolve that tension, and I'm not sure it can from the outside.
Mimi
What it does show is that Anthropic is genuinely deliberating. They're not just reflexively launching. They're evaluating benchmarks, weighing the decision. That's at least consistent with taking safety seriously, even if the outcome might be to launch anyway.
O Pulso
- OpenAI's GPT-6 Astra, deployed earlier this month, has already captured 13% of tracked enterprise AI spending versus Anthropic's 8%, creating a measurable and growing revenue gap.
- The tension is sharpened by CEO Dario Amodei's own recent 3,800-word essay urging the entire AI industry to slow down — a call publicly endorsed by Sam Altman and Elon Musk, now awkwardly shadowed by Anthropic's internal push to accelerate.
- Anthropic's annualized revenue has surged past $65 billion, dwarfing OpenAI's $40 billion run rate, suggesting the competitive threat is real but not yet existential — though investors and insiders are watching closely.
- Structural pressures compound the OpenAI rivalry: cheaper open-source Chinese alternatives are undercutting premium pricing, and key enterprise customer Meta Platforms is actively reducing its dependence on Claude.
- Internal teams are still evaluating safety benchmarks for the next system, and no launch decision has been made — leaving the company suspended between its public principles and its private calculations.
In the months before a closely watched public offering, Anthropic finds itself at a crossroads that few companies are forced to navigate so openly: its chief executive has just warned the world that frontier AI development moves too fast, and yet the market is moving faster still. OpenAI's latest model has begun pulling enterprise spending away from Anthropic's own flagship system, and the company is now quietly debating whether to release a new model of its own. The decision, still unresolved, will reveal whether safety advocacy and competitive survival can coexist — or whether, when the stakes are high enough, one always yields to the other.
Anthropic is wrestling with a decision that cuts to the heart of its public identity. Just two weeks after CEO Dario Amodei published a sweeping essay calling on the AI industry to slow frontier model development — warning of autonomous agents capable of hacking systems without human oversight — the company is internally debating whether to release a new AI model of its own. The timing is not coincidental.
OpenAI deployed GPT-6 Astra on September 3, and the move has already shifted enterprise spending patterns. Data from corporate expense platform Ramp shows Astra capturing roughly 13 percent of monitored enterprise AI expenditures, compared to about 8 percent for Anthropic's Claude Fable. In a market where every percentage point represents millions in revenue — and with an IPO on the horizon — that gap is difficult to ignore.
The commercial pressure, however, may be less dire than the raw numbers suggest. Anthropic's annualized revenue had already climbed past $65 billion by late July, a sevenfold increase from the end of 2025, while OpenAI's rate crossed $40 billion in the same period. Several investors preparing capital for both companies' IPOs argue that Astra does not pose an existential threat. Enterprise software moves slowly, and switching costs are high.
Still, the competitive landscape extends well beyond OpenAI. Open-source alternatives from China are offering businesses cheaper paths to internal AI infrastructure, and Meta Platforms — historically one of Anthropic's largest enterprise accounts — is actively scaling back its reliance on Claude as it expands its own computing models.
What makes this moment distinctive is the collision between Amodei's public advocacy and the company's private calculations. He did not call for a pause in Anthropic's own work — he called on the entire ecosystem to move more deliberately. Yet the company now finds itself weighing whether to accelerate its own release schedule in direct response to a rival's move. The internal teams are still running safety benchmarks, still deliberating. The answer, and what it reveals about the relationship between principle and pressure, is not yet written.
Anthropic is wrestling with a decision that cuts to the heart of its public identity. Two weeks after CEO Dario Amodei published a sweeping 3,800-word essay calling on the entire AI industry to pump the brakes on frontier model development—warning of scenarios where autonomous agents could hack systems without human intervention—the company is now internally debating whether to release its own new AI model. The timing is not accidental. OpenAI deployed GPT-6 Astra on September 3, and the move has already shifted enterprise spending patterns. According to data from corporate expense platform Ramp, Astra is capturing roughly 13 percent of monitored enterprise AI expenditures, compared to about 8 percent for Anthropic's Claude Fable. That gap matters in a market where every percentage point represents millions in revenue.
The internal discussions, described by three people with knowledge of the deliberations, reveal a company caught between two competing imperatives. On one side sits the market: OpenAI's momentum is real, and Anthropic's anticipated initial public offering looms. On the other sits the public stance Amodei just articulated. His essay, which drew public endorsement from OpenAI's Sam Altman and Elon Musk, framed rapid AI development as a systemic risk. Now Anthropic's teams are evaluating the safety benchmarks of their next system while grappling with whether to greenlight a launch at all.
The commercial pressure is substantial but perhaps not as dire as the raw spending numbers suggest. Several investors preparing capital for both companies' upcoming IPOs argue that Astra does not pose an existential threat to Anthropic. The enterprise software market moves slowly—replacing entrenched systems takes time, and switching costs are high. More tellingly, Anthropic's annualized revenue run rate had already climbed past $65 billion by late July, a sevenfold jump from roughly $9 billion at the end of 2025. OpenAI's annualized rate crossed $40 billion in July. By that measure, Anthropic is still winning.
But the competitive landscape extends beyond OpenAI. Open-source and open-weight alternatives from China are offering businesses cheaper pathways to build internal AI infrastructure without paying premium token fees to proprietary providers. And Anthropic's own customers are hedging their bets. Meta Platforms, historically one of Anthropic's top enterprise accounts, is actively scaling back its reliance on Claude systems as it expands its own internal computing models. The company faces structural headwinds across the industry, not just a single rival.
What makes this moment distinctive is the collision between Amodei's public advocacy and the company's private calculations. He did not argue for a pause in his own work—he called for the entire ecosystem to move more deliberately. Yet Anthropic now finds itself evaluating whether to accelerate its own release schedule in response to OpenAI's move. The decision will say something about whether the safety warnings were a genuine assessment of risk or a rhetorical position that yields when the market demands it. The company's internal teams are still running the numbers on safety benchmarks, still deliberating. The answer is not yet written.
Citações Notáveis
Amodei warned of dark scenarios involving swarms of autonomous AI agents hacking websites without human oversight— Anthropic CEO Dario Amodei, in his essay on AI safety risks
Quer a matéria completa?
Leia o original em The Times of India ↗
Technology