OpenAI and Anthropic caution about the dangers of AI as they compete to go public.

      TL;DR: OpenAI and Anthropic have recently published papers cautioning that frontier AI is progressing more swiftly than regulations can address. During this time, they also unveiled their most advanced models, provided free developer tools to encourage usage, and submitted confidential paperwork for public offerings.

      In the last two weeks, the world’s leading AI laboratories have released research, blog posts, and policy initiatives highlighting that the rapid advancements in artificial intelligence are becoming increasingly uncontrollable. Concurrently, both have filed confidential documents to pursue public offerings.

      The contradiction is evident. While OpenAI and Anthropic raise alarms about the risks of rapid AI development, they simultaneously fuel growth through new model releases, free promotions, and IPO filings, which would shift them into publicly traded entities driven by the need for accelerated growth.

      The warnings

      Earlier this week, Anthropic released a paper titled “When AI builds itself,” advocating for a coordinated “slowdown or pause” in the development of frontier models across nations. Authored by Marina Favaro and Jack Clark, the paper posits that AI systems are nearing a stage of recursive self-improvement, at which point human oversight may diminish significantly.

      “Absent a global coordination mechanism, companies and governments will face challenging safety decisions amid competitive and geopolitical pressures,” stated Anthropic. The company noted that by May 2026, over 80% of the code integrated into its systems was written by its AI model, Claude, rather than by human developers.

      On Wednesday, Anthropic CEO Dario Amodei published a blog post titled “Policy on the AI Exponential,” asserting that AI is advancing at a “lightning pace” while policy is “lagging significantly.” He advocated for binding regulations, declaring that “the risks are evidently present” and that mere transparency is no longer sufficient.

      OpenAI expressed similar sentiments. On Monday, CEO Sam Altman and chief scientist Jakub Pachocki issued a blog post titled “Built to benefit everyone: our plan,” suggesting the establishment of an “international organisation to coordinate leading AI efforts to mitigate catastrophic risks.”

      They emphasized that such an entity should possess the authority to regulate the pace of frontier AI development to ensure that “societal resilience, safety, and alignment keep pace.”

      Both firms are not alone in expressing concern. An internal dispute within the White House regarding who should regulate AI has stalled federal policy, creating a gap that neither organization appears willing to wait for Washington to address.

      The acceleration

      However, the warnings seem to contrast with the actions taken by both companies. On Tuesday, Anthropic launched Claude Fable 5, a “Mythos-class” model which they described as their most capable publicly available model to date.

      This model is state-of-the-art across nearly all assessed benchmarks, excelling in software engineering, knowledge work, vision, and scientific research. It includes safeguards that divert sensitive cybersecurity and distillation requests to a less advanced model, Claude Opus 4.8, with the safeguards activating in less than 5% of sessions.

      The company did not respond to inquiries regarding the apparent contradiction between its risk alerts and its model releases.

      OpenAI also introduced GPT-5.5 in late April, billing it as the “smartest and most intuitive” model they have developed. This model set new benchmarks in agentic coding, computer utilization, and knowledge work.

      OpenAI representatives did not respond to requests for comments either.

      Moreover, both laboratories are promoting rapid adoption through complimentary usage incentives. Anthropic has increased Claude Code weekly limits by 50% for paid subscribers through mid-July, while OpenAI provided enterprise customers with two months of free Codex access.

      These strategies aim to lock developers into each company’s ecosystem in anticipation of what both regard as a pivotal year for AI tools.

      The IPOs

      Perhaps the most striking example of this contradiction is the race for IPOs. Anthropic confidentially filed its S-1 registration statement with the SEC on June 1, shortly after securing a $65 billion Series H round that reportedly valued the company at $965 billion.

      OpenAI followed on June 8 by submitting its own confidential S-1 with Goldman Sachs and Morgan Stanley acting as lead underwriters. The last private funding round valued OpenAI at about $852 billion, and analysts speculate that a public listing could push the valuation past $1 trillion.

      Going public would subject both firms to the pressures of quarterly earnings, a factor that historically drives tech companies toward rapid growth at any cost. It is challenging to reconcile calls for a coordinated global slowdown in AI development with the demands of a publicly traded company required to release new products and meet revenue targets every three months.

      The irony is significant

      There is, undoubtedly, a more generous perspective to consider. Both labs may genuinely recognize the risks and could be using their policy papers to advocate for regulations that would apply to all players in the field.

      From this viewpoint, the warnings serve not as hypocrisy but as a strategy: to innovate quickly, sound the