OpenAI and Anthropic have spent the past two weeks publishing papers warning that frontier AI is advancing faster than regulation can keep up. In the same stretch, both released their most powerful models yet, offered free developer tools to drive adoption, and filed confidential S-1 paperwork to go public.TL;DR
In the past fortnight, the two most powerful AI labs on the planet have published research papers, blog posts, and policy proposals warning that frontier artificial intelligence is advancing faster than anyone can control it. In the very same stretch, both filed confidential paperwork to go public.
The tension is hard to miss. OpenAI and Anthropic are simultaneously sounding the alarm on the dangers of rapid AI development and stoking the fire with new model launches, free usage promotions, and IPO filings that would turn them into publicly traded companies under pressure to grow even faster.
Last week, Anthropic published a paper titled “When AI builds itself,” calling for a coordinated “slowdown or pause” in frontier model development across countries. The paper, authored by Marina Favaro and Jack Clark, argued that AI systems are approaching recursive self-improvement, a point at which humans lose meaningful oversight of the development process.
“Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures,” Anthropic wrote. The company disclosed that as of May 2026, more than 80% of code merged into its own codebase was written by Claude, its AI model, not by human engineers.
On Wednesday, Anthropic CEO Dario Amodei published a blog post titled “Policy on the AI Exponential,” arguing that AI is moving at a “lightning pace” while policy is “moving very slowly.” He called for binding regulation, writing that “the risks are clearly here” and that transparency alone is no longer enough.
OpenAI struck a similar chord. On Monday, CEO Sam Altman and chief scientist Jakub Pachocki published a blog post titled “Built to benefit everyone: our plan,” proposing an “international organisation that helps coordinate leading AI efforts to reduce catastrophic risk.”
The pair said this body should have the power to slow frontier AI development so that “societal resilience, safety, and alignment can keep pace.”
The two companies are not alone in their concern. A White House internal fight over who gets to regulate AI has stalled federal policy, leaving a vacuum that neither lab seems willing to wait for Washington to fill.
But the warnings sit awkwardly alongside what both companies have actually been doing. On Tuesday, Anthropic released Claude Fable 5, a “Mythos-class” model that the company described as its most capable ever made publicly available.
It is state-of-the-art on nearly all tested benchmarks, excelling in software engineering, knowledge work, vision, and scientific research.
The model includes guardrails that route sensitive cybersecurity and distillation requests to a less capable model, Claude Opus 4.8. Anthropic said the safeguards trigger in fewer than 5% of sessions.
Representatives for the company did not respond to requests for comment on the apparent tension between its risk warnings and model launches.
OpenAI released GPT-5.5 in late April, calling it the “smartest and most intuitive” model it has ever built. The model set new benchmarks in agentic coding, computer use, and knowledge work.
Representatives for OpenAI did not respond to requests for comment.
Both labs are also encouraging rapid adoption through free usage perks. Anthropic raised Claude Code weekly limits by 50% through mid-July for paid subscribers, while OpenAI offered enterprise customers two months of free Codex access.
The moves are designed to lock developers into each company’s ecosystem ahead of what both expect to be a decisive year for AI tooling.
Perhaps the starkest illustration of the contradiction is the IPO race. Anthropic confidentially submitted its S-1 registration statement to the SEC on 1 June, roughly a week after closing a $65 billion Series H round that valued the company at reportedly $965 billion.
OpenAI followed suit on 8 June, filing its own confidential S-1 with Goldman Sachs and Morgan Stanley as lead underwriters. The company’s last private round valued it at reportedly $852 billion, with analysts suggesting a public listing could push the valuation past $1 trillion.
Going public would subject both companies to quarterly earnings pressure, a force that historically pushes technology firms toward growth at any cost. It is difficult to reconcile calls for a coordinated global slowdown in AI development with the mechanics of a publicly traded company expected to ship new products and hit revenue targets every three months.
There is, of course, a charitable reading. Both labs may genuinely believe the risks are real and may be using their policy papers to advocate for guardrails that would apply equally to competitors.
In that framing, the warnings are not hypocrisy but strategy: build fast, warn loudly, and hope regulation arrives before the technology outpaces it.
But the less charitable reading is also plausible. Publishing safety research and calling for international coordination costs nothing.
Filing for an IPO, launching frontier models, and handing out free developer tools is where the actual commitments lie. The policy papers may serve as reputational insurance, a way to say “we warned you” if things go wrong, without any obligation to slow down unilaterally.
The truth likely sits somewhere in between. What is clear is that the gap between what these companies say and what they do is widening, and neither the market nor regulators seem to be demanding they close it.
Get the most important tech news in your inbox each week.
