Over the past fortnight the AI world has witnessed a striking paradox. OpenAI and Anthropic, the two most powerful research labs on the planet, issued stark warnings that frontier artificial intelligence is developing faster than governments can regulate, yet they also rolled out their most capable models to date and moved toward public listings.

Anthropic’s warning began with a research paper titled “When AI builds itself.” Co‑authors Marina Favaro and Jack Clark called for a coordinated global slowdown or pause on frontier model development, warning that recursive self‑improvement could soon outstrip human oversight. The paper noted that, as of May 2026, more than 80 % of code merged into Anthropic’s codebase was generated by its own Claude model.

CEO Dario Amodei reinforced the message in a blog post, "Policy on the AI Exponential," arguing that policy is lagging behind a "lightning‑pace" AI surge and urging binding regulation. OpenAI’s leadership echoed the sentiment on Monday. CEO Sam Altman and chief scientist Jakub Pachocki published a joint blog entry, "Built to benefit everyone: our plan," proposing an international organization empowered to slow frontier AI development and keep safety and alignment in step with progress.

New model launches

Even as the warnings circulated, Anthropic introduced Claude Fable 5, described as a "Mythos‑class" model and the company’s most capable system ever released publicly. The model set new records on benchmarks ranging from software engineering to scientific research. It includes guardrails that divert sensitive cybersecurity or distillation requests to a less powerful companion, Claude Opus 4.8, with safeguards activating in fewer than five percent of sessions.

OpenAI, for its part, unveiled GPT‑5.5 in late April, branding it the "smartest and most intuitive" model it has built. The system broke performance barriers in agentic coding, computer‑use tasks, and knowledge work. Both firms paired the launches with incentives designed to lock developers into their ecosystems: Anthropic lifted Claude Code usage limits by 50 % for paid subscribers through mid‑July, while OpenAI granted enterprise customers two months of free Codex access.

IPO filings raise stakes

The most tangible sign of a shift toward relentless growth came with confidential S‑1 filings. Anthropic submitted its registration statement to the SEC on June 1, a week after closing a $65 billion Series H round that reportedly valued the company at $965 billion. OpenAI followed on June 8, filing its own S‑1 with Goldman Sachs and Morgan Stanley as lead underwriters. The latest private round placed OpenAI’s valuation at roughly $852 billion, fueling speculation that a public debut could push the market cap past the trillion‑dollar mark.

Going public would subject both companies to quarterly earnings pressure, a dynamic that historically pushes tech firms to prioritize rapid growth over caution. Critics argue that the contrast between the labs’ public safety pleas and the profit‑driven mechanics of a listed company deepens the regulatory vacuum that the White House’s internal AI policy debate has left unfilled.

Supporters suggest the labs are simply using their research and policy outreach to shape a regulatory framework that could eventually apply to all competitors, thereby protecting their own market position while still advancing technology. Skeptics, however, view the safety papers as low‑cost reputation insurance—a way to say "we warned you" without committing to slow development.

The tension between caution and expansion is now unmistakable. As OpenAI and Anthropic continue to pump out more powerful models, offer free tools to developers, and chase trillion‑dollar valuations, regulators and the public will watch closely to see whether the industry’s self‑policing can keep pace with its own ambitions.

Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.