Anthropic Faces Silicon Valley Backlash Over Competition and Open-Weight AI Stance

A growing number of Silicon Valley founders, software executives, and AI researchers are souring on Anthropic, citing competitive product launches that stepped on partner businesses, changes to data retention policies, and what critics call self-serving warnings about open-weight AI models. The tension became public after Anthropic released Claude Design in April 2026, which Figma CEO Dylan Field said came without adequate warning. CEO Dario Amodei responded Monday with a letter clarifying his position, but skepticism about Anthropic's motives is deepening as the company eyes a potential IPO this fall.

What happened

Detail	Fact
Anthropic founded	2021, by Dario Amodei and six other former OpenAI employees
Disputed product launch	Claude Design, released April 2026
Partner affected	Figma, an Anthropic partner, whose CEO Dylan Field criticized the launch
Amodei letter published	Monday (July 28, 2026)
Planned IPO window	As soon as fall 2026

Founders and researchers who spoke to The Wall Street Journal said they began moving to cheaper AI alternatives after watching Anthropic ship tools that competed directly with companies it had previously partnered with. The clearest example was Claude Design, released in April. Figma CEO Dylan Field, whose design platform is an Anthropic partner, told an event audience that Anthropic had not been “consistently candid in their communications.”

Sarah Sachs, head of AI at Notion and an attendee of that event, put it plainly: “Some companies might have been overly trusting.”

What Amodei actually said about open-weight models

A separate strand of criticism targets Anthropic’s public position on open-weight AI models. These are models whose weights (the internal parameters that define how a model thinks) are publicly released, allowing anyone to run and modify them. They are typically far cheaper than hosted models like Claude.

Dario Amodei and other Anthropic executives have argued repeatedly that open-weight models are harder to control and pose security risks, especially as capabilities advance. After a group of major US companies including Nvidia and Microsoft signed a letter supporting open models, Amodei published his own letter on Monday to clarify his stance.

“Anthropic has never advocated for a ban on open-weights models,” he wrote. “Open-weights models that don’t have dangerous capabilities are a public good.” His specific concern is authoritarian governments, particularly China, using open-weight AI to gain military superiority or enable mass repression.

Amodei called on policymakers to crack down on “industrial-scale distillation,” a process where a model is queried millions of times so its outputs can train a competing model. Both Anthropic and OpenAI have accused some Chinese model-makers of doing exactly this. The Trump administration is reportedly considering adding certain Chinese AI companies to a trade blacklist.

Why it matters

The criticism arrives at a sensitive moment. Anthropic is reportedly planning to go public as soon as fall 2026. Critics, including some Trump administration officials, argue the company’s warnings about open-weight risks conveniently serve that financial goal: cheaper open-weight models, several of them from China, are the most direct competitive threat to Anthropic’s revenue.

Vishal Misra, vice dean of computing and AI at Columbia University’s engineering school, told the Journal that while Anthropic builds strong products, its public statements on AI risk “don’t hold up to scrutiny.” He argued the company generates public fear to gain regulatory leverage.

The distillation argument also cuts both ways. Mark Suman, co-founder and CEO of AI productivity startup Maple, pointed out: “It’s interesting they’re against distillation when they distill the entire Internet without paying royalties.” His point: training large models on scraped web data is not entirely different in kind from the distillation practices Anthropic wants banned.

Not everyone agrees with the critics. Kevin Bryan, an associate professor at the University of Toronto specializing in innovation, defended the underlying safety concern. “Anything that is open source is immediately jailbroken,” he said. “Once the model’s out, that’s it.” You can follow the broader debate over open-source AI labels and security claims in our earlier coverage of the David Sacks vs Anthropic dispute.

Our take

Anthropic is caught in a real contradiction and it is not entirely of its own making. The company was founded on safety principles, publishes serious research, and has genuine reasons to worry about capable AI in adversarial hands. But it is also a well-funded commercial operation heading toward an IPO, and those two identities are now visibly in tension.

The Figma situation is the more damaging story. Safety philosophy is debatable. Blindsiding a partner with a competing product launch is a trust problem, and trust is hard to rebuild. If Anthropic wants developers to build on Claude, it needs a clearer policy on what spaces it will and will not enter. Right now, founders don’t have that assurance.

For businesses currently evaluating which AI platform to build on, this is a real consideration. Vendor lock-in risk is not just about pricing or model quality. It is about whether your AI provider might ship a product next quarter that competes with yours. Our AI integration work with clients increasingly involves structuring workflows so they are not entirely dependent on a single model provider for exactly this reason.

What to do about it

1. Audit which parts of your product or workflow rely exclusively on Anthropic’s Claude API and identify where a swap to an alternative model would be practical.
2. Review Anthropic’s current data retention policy if you handle sensitive customer data, since the company recently changed those terms.
3. Watch the Trump administration’s moves on Chinese AI company blacklisting. If major open-weight models from China become restricted for US businesses, your tooling options narrow quickly.
4. If you are evaluating open-weight models as a cost-saving measure, talk to a technical team about what guardrails you would need to add, since hosted models include safety layers that open-weight deployments require you to build yourself.

The safest position right now: treat any single AI provider as a dependency to be managed, not a permanent partner.

Source: Bing News · Claude AI