• Palantir co-founder Joe Lonsdale warns that OpenAI and Anthropic’s push for frontier AI safety standards could entrench incumbents and stifle competition.
  • The dispute centers on mechanism: formal pre-approval regulation versus ex post liability for demonstrable harms.
  • Palantir’s robust commercial growth underscores its stake in a permissive AI deployment environment.

Joe Lonsdale, co-founder of Palantir Technologies (PLTR), has sharply criticized OpenAI and Anthropic’s efforts to shape AI policy, calling their push for frontier safety standards “very dangerous” and warning it could create a burdensome regulatory regime that favors well-funded incumbents.

The disagreement is not over whether AI companies should bear responsibility for harms, but over how to achieve that. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have both recently endorsed the principle of “pacing the frontier” of advanced AI development. Amodei proposed independent evaluators with deep access to company systems and greater coordination among labs and governments. Altman publicly backed that approach and said OpenAI would adopt independent evaluators with employee-like access. Reporting also indicates OpenAI is supporting consistent national frontier-AI safety requirements.

Lonsdale rejects the existential-risk framing that underpins those proposals. In recent media appearances, he said he does not see an AI “existential risk,” arguing that panic is being used to justify new regulation. Instead, he favors clear, severe liability for demonstrable harms rather than broad pre-approval regulation. His central criticism is that companies with extensive legal and policy teams may benefit when rules become complex, because potential challengers must spend heavily to comply—a classic regulatory-capture concern.

“What institutional investors like us are really focused on is regulatory stability,” Lonsdale said, according to people familiar with his thinking. “But stability shouldn’t mean a licensing system that locks in today’s winners.”

The debate arrives amid a powerful investment cycle in AI models, data centers, chips, and enterprise implementation. A stringent licensing or testing regime could slow deployments and add fixed compliance costs. Lonsdale argues those costs would be easier for OpenAI, Anthropic, Google (GOOG), and Microsoft (MSFT) to bear than for startups.

Palantir’s own financial performance illustrates the commercial stakes. In its most recently reported quarter, Q2 2026, the company posted revenue of $1.935 billion, up 93% year over year. U.S. commercial revenue grew 149% to $764 million. GAAP net income was about $1.062 billion, a 55% margin, and the company ended the quarter with roughly $9.2 billion in cash, equivalents, and short-term U.S. Treasuries. Management lifted full-year 2026 revenue guidance to approximately $8.15–$8.16 billion and forecast Q3 revenue of about $2.16 billion.

Those figures underline why Palantir has a strong commercial stake in a policy environment that speeds AI deployment, particularly in U.S. enterprise and government markets. Palantir is a major public AI-software vendor—its platforms include Gotham, Foundry, Apollo, and AIP—rather than a frontier-model developer like OpenAI or Anthropic.

The policy dispute maps onto a broader U.S. political question: should Washington establish a uniform federal AI framework, or should AI safety be handled largely through ordinary liability law and market competition? The Trump administration’s AI policy has generally emphasized U.S. technological leadership, deregulation, and reducing obstacles to private-sector AI development. The White House’s 2026 national framework favors a minimally burdensome federal standard and seeks to preempt certain state AI laws, while preserving state authority in areas including child safety and data-center infrastructure.

The United States still lacks a single comprehensive federal law governing AI development and use. That gap is central to the dispute. OpenAI and Anthropic are advocating national consistency with frontier-model safety obligations, while Lonsdale argues that a centralized regulator could become slow, politically influenced, and captured by major firms.

The debate also has international implications. Amodei’s position is not a call for unilateral U.S. restraint; it seeks coordinated pacing and safety measures without surrendering America’s strategic lead. Lonsdale stresses that restrictions could weaken U.S. competitiveness and security if rivals continue advancing.

Industry observers note the unusual alignment of competitors Altman, Amodei, and Elon Musk behind at least some form of slower frontier progress. That has made Lonsdale’s deregulatory and liability-first stance a prominent counterargument rather than a fringe view.

The most likely policy landing zone, according to analysts, is neither a complete pause nor a wholly unregulated market. A compromise could combine narrow federal rules for the highest-risk capabilities, mandatory evaluations at defined thresholds, incident reporting, and liability standards tailored to negligence and product design, while preserving latitude for lower-risk enterprise AI.

The key unresolved question is whether the United States can make those safeguards credible without creating a de facto licensing system that favors the largest model developers. That is precisely the tension at the heart of Lonsdale’s criticism.

Palantir declined to comment beyond Lonsdale’s public remarks. OpenAI and Anthropic did not respond to requests for comment.

Correction: An earlier version of this article misstated the quarter in which Palantir reported $1.935 billion in revenue. It was Q2 2026, not Q2 2025.