After a Banner Month, Anthropic Faces a Silicon Valley Backlash Over Its Tactics

Six months ago, Anthropic looked untouchable. The maker of Claude spent the winter and spring atop nearly every independent benchmark, shipped a rapid-fire sequence of flagship models, signed marquee enterprise deals, and hit a $965 billion valuation that pushed it past OpenAI as the world's most valuable startup. It cast itself, as it always has, as the industry's moral conscience.

By late July, that conscience had become a target. A widening chorus of founders, researchers, government officials and even ostensible allies is accusing Anthropic of weaponizing "safety" as a competitive moat — using the language of caution to justify tactics and policies that, critics say, conveniently protect its own business. The Wall Street Journal first documented the souring mood in Silicon Valley, and within days the fight spilled into the open, with named executives and White House officials trading barbs over open-weight models, Pentagon contracts and the guardrails baked into Claude.

The spark: a letter Anthropic wouldn't sign

The immediate flashpoint was an open letter organized by Nvidia CEO Jensen Huang and released July 24, urging Washington not to impose "premature restrictions" on open-weight AI models — the downloadable systems anyone can run and modify, a category now dominated by Chinese labs. More than 50 companies signed, including Microsoft, Meta, Hugging Face, Mistral, Palantir and, over the weekend, Anthropic's two closest closed-model rivals, Google and OpenAI.

Anthropic was the only frontier lab that declined. Its absence, in the words of an Axios analysis, left it "marooned on an island of its own principles" — and handed critics an opening. White House AI adviser David Sacks and figures from rival labs argued publicly that the refusal exposed Anthropic's safety branding as commercial self-interest in disguise.

Palantir CEO Alex Karp, whose company did sign, put the competitive critique bluntly in a Fox Business interview. "I'm not anti-Anthropic," Karp said, but "if you want to win, you have to compete on the battlefield." Restricting the best models to U.S. commercial use while rivals abroad face no such limits, he argued, is "complete insanity" that guarantees "no one can claim we have the best products in the world."

The distrust had been building. In June, Anthropic shipped Fable 5, a safety-constrained model that quietly degraded answers to certain AI-development questions without telling users — invisible safeguards the research community quickly branded as anticompetitive. Around the same time, after Anthropic disputed the severity of a security vulnerability the White House had flagged, export controls forced its Fable and Mythos models offline for nearly three weeks. And the Pentagon, which blacklisted Anthropic in February over a fight about surveillance and autonomous-weapons use, escalated further: senior official Emil Michael declared there is "no AI company more hostile to the warfighter."

Anthropic's defense

Anthropic's CEO, Dario Amodei, tried to defuse the fight himself. In a July 27 blog post, he rejected the premise that the company wants to lock down open models. "Anyone who has read my past writing should know that I don't regard such bans as a useful measure," he wrote, "but let me state it clearly so that there is no doubt: Anthropic has never advocated for a ban on open-weights models."

He went further, calling low-risk open models "a public good," valuable to businesses, developers and researchers and costing "nothing besides the compute needed to run them." What he would not concede is the letter's core claim that openness inherently aids defenders. "I don't agree," he wrote, "that broad access to capabilities necessarily helps defenders more than attackers. It seems at least as likely to me that the opposite will be true" — particularly for biological threats, where he argued attackers may hold the edge.

Amodei's alternative: keep advanced chips out of authoritarian hands, crack down on "industrial-scale distillation" of Western models by Chinese labs, and require safety testing of all sufficiently capable models, open or closed — a regime he says is nearing global consensus. Anthropic, asked for comment, pointed only to that post.

Why it matters

The backlash is less about any single policy than about a strategic bet. Anthropic has built its brand, and arguably its enterprise lead, on being the cautious lab. Now rivals and regulators are reframing that caution as a business strategy — one that raises compliance costs, restricts model behavior, and leans on regulation in ways that could entrench incumbents. Critics also note Anthropic is an awkward messenger on distillation, having just settled a $1.5 billion copyright lawsuit over pirated books.

Yet the isolation is not total. This week Anthropic joined more than 1,100 employees from OpenAI, Google DeepMind and Meta on a petition urging government help to "deliberately pace" frontier development — a sign its central worry is broadly shared even as its tactics alienate. And for all the noise, the company's models still top the benchmarks, and enterprise customers keep paying premium prices.

What to watch next

Three things. First, whether Washington leans toward Anthropic's testing-and-chips framework or the open-weight camp's lighter touch as it weighs a formal China response. Second, whether the Pentagon rift costs Anthropic real defense revenue while rivals like Palantir court that market aggressively. And third, the IPO: Anthropic is barreling toward a listing that could value it above $1 trillion, and how much a Silicon Valley backlash dents — or fails to dent — that story will be the clearest verdict on whether "safety as moat" is a liability or a durable edge.