Anthropic was founded on the promise of building safer, more ethical AI. But now it finds itself facing a growing revolt across Silicon Valley.
Once celebrated as the high-minded alternative to OpenAI, the startup now finds itself accused of regulatory capture, product cannibalization, and hypocrisy as its commercial strategies collide with the broader tech ecosystem.
The backlash represents a critical turning point for the startup, exposing deep fissures between proprietary frontier AI labs and the broader developer ecosystem. What began as quiet frustration among software partners has erupted into a full-scale policy and commercial battle over the future of artificial intelligence (AI).
“I’d argue the backlash is more about the industry waking up to the fact that model providers are competitors, not utilities,” LevelAI CEO Ashish Nagar said. “Every company that built its product or its AI strategy on a single lab’s API is now watching labs move up the stack into their markets. Enterprises shouldn’t just be focused on finding a more trustworthy frontier lab. The lesson here is that every company needs to build an architecture where any model is a replaceable component.”
The Ecosystem Strikes Back
The trouble began in the market. Founders and software executives report turning to cheaper, open-weight AI alternatives after Anthropic began launching products that directly competed with its own business partners.
The tipping point came with the release of Claude Design, a tool seen as a direct shot across the bow of partner Figma. According to attendees at an industry event shortly after, Figma CEO Dylan Field remarked that Anthropic hadn’t been “consistently candid” in its communications. The incident sent shockwaves through the startup community. “Some companies might have been overly trusting,” noted Sarah Sachs, head of AI at Notion.
Distrust deepened when Anthropic launched Fable 5, a restricted iteration of its Mythos model.
The system initially degraded answers to AI development queries without informing users — a hidden safety guardrail that researchers blasted as anticompetitive. Coupled with a controversial 30-day data retention policy that spooked enterprise clients handling sensitive information, many startups concluded that relying on Anthropic posed an existential threat to their businesses.
“The frontier labs are hoarding intelligence,” said AI researcher Mahesh Lambe during a pro-open-weights protest in San Francisco. “It needs to be democratized.”
“Anthropic’s challenge is not simply that it is moving up the stack. It is that the company risks entering markets without fully understanding how the people in those markets actually work,” said Matt Darrow, CEO of Vivun. “In sales, for example, Anthropic’s approach still appears heavily shaped by an engineering mindset. Most sellers do not want to become AI builders. They want technology that understands their business, fits into their workflow. and helps them perform in the moment that matters. The backlash is what happens when a company presents itself as a neutral foundation for an ecosystem, then begins competing with that ecosystem without demonstrating the same depth of domain understanding.”
The Regulatory Shield
As developer trust eroded, critics began questioning Anthropic’s high-profile advocacy for AI safety, framing it as a calculated attempt to erect regulatory moats around its proprietary tech.
CEO Dario Amodei, who co-founded Anthropic in 2021 after leaving OpenAI over safety disagreements, has consistently warned about the risks of open-weight models, which allow users to download and customize AI systems. Critics contend these warnings are designed to stifle open-source competitors, particularly as Chinese labs like Moonshot AI release models that approach frontier U.S. capabilities at a fraction of the cost.
Academia has joined the chorus of skeptics. “They create a scare in the general public around these models,” said Vishal Misra, vice dean of computing and AI at Columbia University. Public fear, he added, gives the company “leverage to get regulations done in their favor.”
Startup founders have also pointed out the apparent contradiction in Anthropic’s opposition to industrial-scale distillation, the practice of using a larger model’s outputs to train a smaller one.
“It’s interesting they’re against distillation when they distill the entire Internet without paying royalties,” said Mark Suman, CEO of startup Maple. “I just think it’s odd that they are now turning around and telling others not to do what they did.”
Amodei’s High-Stakes PR Offense
Facing an isolating narrative, Amodei published a lengthy blog post clarifying his stance, insisting that Anthropic “has never advocated for a ban on open-weights models.”
Instead, Amodei reframed the debate around national security, arguing that the primary threat is authoritarian regimes catching up to U.S. capabilities. He called on Washington to tighten hardware sanctions to prevent advanced semiconductor smuggling to China; clamp down on distillation, citing an instance where China’s Alibaba allegedly executed a massive distillation attack against Anthropic models; and mandate safety testing for all sufficiently powerful models, regardless of whether they are open or closed.
“Protectionist bans would not address my most serious national security concerns,” Amodei wrote, attempting to draw a line between targeted safety protocols and market protectionism.
Amodei’s defensive posture comes as the rest of the tech industry aggressively rallies behind open architecture. Tech giants including NVIDIA Corp., Microsoft Corp., Meta Platforms Inc., and Palantir Technologies Inc. recently signed an open letter calling on Washington to avoid “premature restrictions” on open models. Even chief rival OpenAI added its name to the coalition.
Anthropic stood as the lone major AI lab refusing to endorse the letter, further fueling perceptions that it is out of step with the broader tech community.
“What’s playing out across recent AI agent incidents isn’t really an argument about which model you trust; it’s a warning about trust itself and a real moment for identity access management,” said Ramprakash Ramamoorthy, director of AI research for ManageEngine. “The lesson for every enterprise running AI agents isn’t ‘watch out for the wrong model.’ It’s that any AI agent, including your own, needs an identity boundary that doesn’t waiver. As long as AI agents are managed as tools instead of identities, these blind spots will persist, regardless of which model is powering them.”
With Anthropic eyeing an initial public offering as early as this fall, the rift presents a serious strategic headache. While safety defenders like University of Toronto professor Kevin Bryan argue that fears over open-source risks are “real and widely shared,” Anthropic must now prove to investors and developers that its safety-first mission isn’t just a cover for market dominance.
“Why is anyone surprised? Anthropic isn’t a nonprofit. It’s operating in one of the most competitive and capital-intensive markets in tech,” said Gary Barlet, public sector chief technology officer at Illumio. “What’s happening reflects a broader challenge across the AI industry: Companies may start with ambitious principles around safety and ethics, but as the market matures, commercial pressures become harder to ignore. The real question isn’t whether that tension exists; it’s how companies navigate and communicate the tradeoffs that come with scaling a business.”
