Anthropic’s Self-Writing Code Drives First Profit — and a Plea for a Global Halt

Anthropic reveals 80%+ of its code is now AI-generated, prompting calls for industry slowdown even as it nears $965B valuation, first profit, and IPO.

More than 80 percent of the code in Anthropic’s own development environment is now written by its AI model Claude — a statistic the company itself finds alarming enough to call for a coordinated industry-wide slowdown. The disclosure lands at an awkwardly triumphant moment: Anthropic is on track to post its first-ever operating profit in June 2026, has locked down a $65 billion funding round, and has confidentially filed for an IPO that could come as early as October.

The internal numbers behind the code-writing milestone are stark. Before the launch of Claude Code in February 2025, AI-generated code accounted for only a low single-digit percentage of Anthropic’s output. Today it is north of 80 percent, and engineers are shipping eight times more code per quarter than they did between 2021 and 2024. The model’s success rate on difficult, open-ended tasks has jumped from roughly 15 percent in late 2025 to over 70 percent by the spring of 2026. According to Anthropic, the ability to solve tasks autonomously is doubling about every four months.

That pace of improvement is the engine behind the financial acceleration — and also the reason Anthropic’s policy chief, Jack Clark, told the BBC that fully AI-authored code could be a reality within two years. But the same trajectory prompted Anthropic research head Marina Favaro and Clark to publish a report on June 4 titled When AI Builds Itself, warning that models are approaching the concept of “recursive self-improvement” — the capacity to develop their own capabilities by writing better code. The report explicitly states that this threshold has not been reached and is not inevitable, but it could arrive sooner than regulators or institutions are prepared for.

The financial backdrop makes the warning particularly delicate. Anthropic closed a Series H round in late May led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, bringing its post-money valuation to $965 billion — vaulting it past OpenAI as the world’s most valuable AI startup. Revenue for the second quarter of 2026 is projected at roughly $10.9 billion, more than double the $4.8 billion posted in the first quarter. The annualized run rate now stands at $47 billion, compared with $10 billion at the end of 2025. Operating profit for the June quarter is estimated at $559 million, marking the first time revenue has outpaced the enormous computing and training costs that have kept the company in the red.

To sustain that growth, Anthropic has inked two strategic compute-capacity deals. With SpaceX, it will take the entire output of the Colossus-1 data center — over 300 megawatts of power and 220,000 NVIDIA GPUs. Separately, it has signed a multi-gigawatt agreement with Alphabet for next-generation Tensor Processing Units. On the enterprise side, the Claude Partner Network has deepened its reach: Accenture has trained 30,000 staff on Claude, Cognizant is deploying the technology across roughly 350,000 employees, Deloitte is rolling it out to 470,000, and PwC is introducing Claude Code and Cowork to its U.S. teams.

Against that growth surge, the company’s call for a halt is deliberately multilateral. Anthropic says a pause would require several well-funded frontier labs across multiple countries to stop simultaneously, with clear trigger conditions and oversight by an independent body. A unilateral stop, it argues, would only hand advantages to less cautious players. OpenAI, xAI, Alphabet, Meta, and Mistral have not commented.

The timing has not gone unnoticed by critics. On June 2, President Donald Trump signed an executive order asking AI companies to voluntarily submit frontier models for government review 30 days before release. Anthropic’s own security stance has shifted: in February it reversed a central safety promise, saying it would no longer hold back potentially dangerous AI if competitors were close to matching its capabilities. Some analysts see the safety rhetoric as positioning. One wrote that Anthropic is hyping its capabilities to investors while simultaneously pushing a narrative that its technology is so advanced it requires permanent massive funding — a convenient story for a company heading toward the public markets.

The Anthropic Institute plans to convene policymakers, researchers, and rivals in the coming months to hash out next steps. Whether the broader industry joins the pause will likely shape how investors judge the company’s credibility when its S-1 documents become public this autumn.