Why AI Leaders Have Changed Their Minds About AI Safety, Elon on UHI, Anthropic’s IPO

The White House Super Intelligence Accord favors continued AI development with layered internal and external oversight, while speakers also called for safeguards against cyber, biological, and chemical risks. The accord’s approach to overregulation does not resolve liability for harm caused by AI agents.

Speakers framed AI liability as a trade-off between accountability and U.S. competitiveness: one argued protections should vary with system capability, while another warned that liability for foundation-model companies could push users toward Chinese open-weight models. FTC Chairman Andrew Ferguson, by contrast, argued developers cannot avoid responsibility by treating agents as independent actors.

The discussion linked alignment to both technical capability and governance: one speaker said post-training reshapes how models allocate capabilities and compute, citing a reported 10,000x per-parameter improvement, while others disagreed on whether alignment follows from capability. Evaluating powerful systems such as a swarm that can solve Navier–Stokes may require exceptional computing resources and a pool the speaker estimated at fewer than 100 people worldwide.

Speakers argued that AI-enabled abundance would not ensure broad access on its own: one proposed lowering the cost of essentials such as healthcare, education, and energy, while another warned that ownership of productive infrastructure could concentrate wealth. Proposals included collectively owning robots and data centers, giving citizens agents to coordinate services, and expanding broad equity exposure through mechanisms such as Trump accounts; the latter was presented as an argument, not a confirmed plan.

Anthropic’s reported financial profile combines rapid growth with substantial commitments: its S-1 figures include 2025 revenue of $4.59 billion, up 12x year over year, an $8 billion operating loss, roughly $500 billion in future cloud and infrastructure commitments, and $20 billion in cash. Speakers also questioned the tension between those obligations and the company’s prospectus warnings about severe AI risks, while citing reporting that much of its revenue or revenue growth may depend on AWS-hosted models.

Anthropic’s reported founder-control structure would give co-founders 51% of voting power despite holding 14% of shares, with super-voting shares intended to protect the safety mission from public-shareholder pressure. The discussion noted that this could preserve founder control even if only two of the current seven founders remained, while acknowledging potential conflicts between safety and financial interests.