Anthropic is discovering that it is fast becoming as popular as the Boston Strangler in Silicon Valley.
According to the Wall Street Journal, a backlash is brewing against Anthropic, with startups and researchers muttering about rival products, closed AI and regulatory sermonising.
Startup founders, software executives and AI researchers said they began shifting to cheaper models after Anthropic launched tools that competed with those of other firms.
One of the more awkward launches was Claude Design in April, which some saw as a direct shot at Figma, an Anthropic partner.
Figma chief executive Dylan Field said Anthropic had not been “consistently candid in their communications.”
Notion head of AI Sarah Sachssaid: “The whole industry learned a lot of lessons from how that dynamic played out. Some companies might have been overly trusting.”
Anthropic has attracted more grief for changing its data retention policies and for shipping a powerful model that refuses to answer some AI research questions.
Researchers pushing for transparency were less than thrilled, especially when the outfit’s safeguards arrived with little visible warning.
There is growing scepticism about Anthropic’s warnings about “open weight” AI, in which users can customise models and bolt on their own safeguards.
Several Chinese open-weight models have impressed users recently, with capabilities edging closer to frontier US systems and prices that do not require a VC blood sacrifice.
An Anthropic spokeswoman defended the company’s moves, saying they were needed to keep models safe and stop adversaries creating severe risks.
Anthropic chief executive Dario Amodei and other executives have repeatedly warned about open-weight AI systems, which many researchers see as harder to control.
Some AI executives say those models lack the same security safeguards as leading US systems, which is rather convenient for the leading US systems.
Anthropic publishes stacks of AI safety research and backs government regulation through endorsements and political donations.
Anthropic and OpenAI have accused some Chinese model-makers of copying their technology through “distillation,” which involves training another model by feeding a model with queries.
Some Trump administration officials, including Treasury Secretary Scott Bessent, have raised worries about distillation, although they have stopped short of action against Chinese companies.
OpenAI offers some open-weight models, and Altman said last year that he regretted the company’s previous stance against the tools.
The Trump administration is considering adding Chinese companies to a trade blacklist, which would effectively prevent many US businesses from using the models.
Critics, including some administration officials, see Anthropic’s warnings as an attempt to hobble rivals before a possible public listing as soon as the autumn.
Technologists and policymakers are worried the AI race could become a duopoly, with Anthropic and OpenAI emerging as market gatekeepers.
Those fears were all over a pro-open-weights protest on Saturday in downtown San Francisco, the AI startup world’s spiritual co-working space.
The protest was co-hosted by the chief executive of Hugging Face, an AI company providing open-source tools.
AI researcher Mahesh Lambe, who attended the protest, said: “The frontier labs are hoarding intelligence. It needs to be democratised.”
Founders and startup executives said Claude Design was a wake-up call for Silicon Valley’s more trusting souls.
Some concluded that Anthropic could launch tools that competed with them and nicked their customers.
Distrust grew in June when Anthropic released Fable 5, a safety-constrained version of its powerful Mythos model, with limitations.
At first, the model degraded answers to some AI development questions without telling users what it was doing.
The research community quickly criticised the invisible safeguards, saying the decision looked anticompetitive and made model capabilities harder to assess.
Anthropic apologised and made those safeguards visible to users but said it would still refuse flagged requests.
A company spokeswoman said Anthropic restricted the model’s responses to stop foreign adversaries using them in “ways that create severe safety risks.”
Fable 5 would not support “zero data retention,” meaning Anthropic would keep user data for 30 days for trust and safety purposes.
Companies handling sensitive data were unimpressed. Some firms worry their data could train future models that help Anthropic launch competing products.
