• Joe Lonsdale, Palantir (PLTR) co-founder and 8VC founder, endorsed Dario Amodei’s leadership at Anthropic during a CNBC interview on September 22.
  • The backing crosses a public divide: Lonsdale has downplayed existential AI risks, while Amodei advocates for a coordinated slowdown of frontier-model development.
  • The endorsement comes as Anthropic’s revenue run rate hit $65 billion by end-July, up from $47 billion in May, and competition over AI safety and deployment intensifies.

A Surprising Alliance in AI’s Safety Debate

Joe Lonsdale, the co-founder of Palantir and founder of venture firm 8VC, said he backs Dario Amodei as the right leader for Anthropic, according to a CNBC interview on September 22. The endorsement is notable not because it signals any leadership change—Amodei remains CEO—but because it bridges a widening philosophical gap in the AI industry. Lonsdale has publicly minimized apocalyptic AI scenarios, saying leaders have responsibilities but “the world is going to be alright.” Amodei, meanwhile, has become one of the most prominent advocates for deliberately slowing the pace of frontier-model development until stronger safeguards are in place.

“Dario is the right leader for Anthropic,” Lonsdale said, according to the interview, framing his support as a vote of confidence in Amodei’s stewardship rather than a shift in control. The comment also comes amid a broader debate over whether AI companies should prioritize rapid capability gains or coordinated safety measures.

Safety as a Commercial Differentiator

Amodei’s September essay, “We Must Pace the Frontier,” called for three concrete steps: independent evaluators embedded deeply enough to assess safety practices, coordination among major developers on safety standards, and international cooperation on frontier-AI risks. The proposal has generated unusual alignment and conflict simultaneously. Reuters reported that OpenAI’s Sam Altman and xAI’s Elon Musk supported a coordinated slowdown, while Lonsdale has rejected the apocalyptic framing.

Despite its safety focus, Anthropic remains commercially aggressive. The company recently announced Claude Opus 5.5, which it says reaches the level of its top Fable 5.1 model on most work while costing 40% less to run than its Opus 5 predecessor. Anthropic’s revenue run rate was $65 billion by end-July, up from $47 billion in May and about $9 billion at end-2025, Reuters reported. The company was reportedly valued at $965 billion after a $65 billion Series H round in May, with a potential public listing under discussion. Run rate is not audited annual revenue; it extrapolates current sales.

The Economics of a Slowdown

If frontier labs were to coordinate on safety standards, the impact would shift spending rather than halt it. Reuters Breakingviews argued that greater safety oversight could redirect substantial investment from merely training ever-larger models toward deployment, evaluation, security, and data-center infrastructure. That may benefit cloud providers, cybersecurity firms, testing specialists, and hardware suppliers beyond the current leaders.

Palantir, for its part, demonstrates the deployment-side boom. The company reported Q2 2026 revenue of $1.94 billion, up 93% year over year, and raised its full-year revenue guidance to $8.150–$8.158 billion. Lonsdale spoke as an investor and tech-industry voice, not as Palantir’s operating leader. The company’s government and defense footprint makes his views especially visible in debates over AI in security and intelligence.

Regulatory and Geopolitical Crosscurrents

Amodei’s proposal sits at the intersection of AI safety, competition policy, and national security. He has called for independent testing and international collaboration, and said limited U.S. antitrust exemptions may be needed so rival frontier labs can coordinate narrowly on safety without facing antitrust exposure. The European Union already provides a regulatory benchmark: its AI Act establishes requirements for general-purpose AI models, with key enforcement powers and transparency obligations becoming enforceable on August 2, 2026.

U.S.–China rivalry complicates any global pact. Amodei has supported stronger U.S. chip-export controls on China and enforcement against alleged model distillation. China’s state-backed Global Times criticized the slowdown proposal as a “Cold War playbook,” showing how safety policy can be interpreted abroad as a means of preserving U.S. technological advantage. European firms including Mistral have also challenged Anthropic’s approach, arguing that a slowdown imposed by established U.S. frontier labs could entrench incumbents.

Anthropic has also faced an unusually consequential U.S. government dispute. According to the Congressional Research Service, a federal judge in March temporarily blocked implementation of a designation that would have stopped federal agencies from using Anthropic technology; GSA subsequently restored Anthropic to USAi.gov and its schedule. That episode underscores why leadership credibility and government relations matter for the company.

What to Watch

In the near term, Anthropic will likely face pressure to demonstrate that its safety commitments translate into concrete, verifiable practices—especially independent evaluation, incident reporting, and stronger customer-facing safeguards. The industry is expected to keep releasing increasingly capable systems despite the slowdown discussion, as Anthropic’s Opus 5.5 launch shows. Policy debate may focus on liability, safety-testing standards, government procurement rules, and whether firms can legally collaborate on common frontier-model controls.

Lonsdale’s endorsement may be strategically important precisely because it signals that respect for Amodei’s leadership can extend beyond the strictest AI-safety camp. But it does not erase their philosophical difference: Lonsdale has generally emphasized optimism and accountability for concrete harms, while Amodei argues that rapidly increasing capability itself warrants coordinated restraint. Representatives for Anthropic and Palantir did not immediately respond to requests for comment.

Correction: An earlier version of this article misstated the timing of Amodei’s essay. It was published in September, not earlier in the year.