A growing trust problem is emerging between frontier AI labs and the companies that pay for their models: customers worry that the labs will not only provide models but also convert operational knowledge and data gathered during deployments into competitive advantage.
How the issue surfaced
A few weeks ago at the Lion Forum — a venture-capital event in Hyannis, Massachusetts — Syed Mohiuddin, head of healthcare at Anthropic, was asked the question on many CEOs’ minds: why should they trust the company not to steal their business? Mohiuddin replied, “It’s a fair question. Because we are both a frontier lab that’s building models and a product company that has applications.”
The tension has appeared publicly as well. Palantir CEO Alex Karp, in a chaotic appearance on CNBC last week, said clients are “livid” and complained they are paying for tokens that “create no value… and they’re going to get my IP.”
Why customer support looks like reconnaissance
Engineers that frontier labs deploy to customers to help implement models — often described as forward-deployed engineers — gain deep insights into how banks, consultancies, retailers, manufacturers and other clients operate. From a skeptical perspective, that customer support can resemble reconnaissance, collecting information that could be used to enter those markets.
Anthropic’s moves in legal and design sectors
Anthropic has already rolled out products for law firms and design firms. The reporting characterizes the primary losers as software companies that previously sold those services, citing Harvey (legal) and Figma (design) as examples. That raises the question: after ingesting enough customer data, what stops labs from launching their own operating businesses and directly competing with client firms?
Open-source models and local deployment
At the same time, open-source models — many developed in China but also an expanding set of American alternatives such as Reflection — are quickly narrowing the capability gap at far lower cost. These models can be downloaded and run locally without sending data back to the labs, reducing customers’ exposure.
A business dilemma and public reassurances
Will Wilson, CEO of code-debugging startup Antithesis, told the author that companies must use lab products “to stay in the race, but doing so requires sending all of their IP, and that is a very uncomfortable thing to do with somebody who might be trying to replace you.” Wilson also said he’s “not scared of Anthropic” and is enjoying Fable 5.
AI labs officially deny plans to displace customers. Major players, including OpenAI and Anthropic, publish enterprise terms intended to limit how customer data may be used. Mohiuddin argued that “Claude Code didn’t kill Replit and Cursor,” and framed Anthropic’s aim as raising the baseline of what’s possible rather than acting as a market replacer.
The core economic question
The more fundamental question is not only whether labs will move downstream to compete directly, but whether they can capture greater value by enabling customers — for example, helping Sullivan & Cromwell draft contracts, Gensler produce architectural plans, or Pfizer discover drugs — or by performing those services themselves.
Signals from token usage
Usage metrics show rapid adoption alongside the mistrust. Semafor reporter Rachyl Jones noted that the share of tokens used by China’s DeepSeek models on OpenRouter has doubled in the past six months as the AI lab seeks greater control over its hardware. Separately, Harvey’s co-founder and president announced on X that the number of tokens processed on its legal AI platform has increased 14-fold over the past six months.
Conclusions
The relationship between frontier AI labs and their customers is both strategic and technical: labs are valuable partners because of their capabilities and customer engagement, but long-term trust and management of data capital remain unresolved. Open-source alternatives and surging token volumes complicate decisions about whether the sector moves toward deeper partnerships or toward labs taking on customer-facing market roles.
