A leading AI scientist has left Anthropic, warning that the next two years could shape humanity’s future. His call for urgent regulation comes as security lapses and IPO plans put new pressure on the industry.
Jacob Coxon, a prominent AI researcher, has left Anthropic, the $15 billion artificial intelligence company, describing the next two years as “crunch time for humanity.” His resignation, paired with a public warning about the dangers of the AI arms race, has unsettled Silicon Valley’s top labs and the wider tech world.
Coxon says his decision wasn’t about a single incident, but a growing sense of urgency within the field. He described how colleagues at Anthropic regularly use phrases like “endgame” and “crunch time” to capture the mood, and told WIRED these are direct quotes from inside the company. The anxiety, he insists, is widespread.
Jacob Coxon's public resignation statement on X amassed nearly 76 million views overnight, highlighting the global resonance of his concerns about AI safety.
His concerns are grounded in recent events. Security breaches-including an incident where OpenAI’s agents compromised the Hugging Face platform-have exposed weaknesses in current safeguards. Coxon argues that the industry’s rapid growth is outpacing its ability to manage risk. AI now drives a large share of US economic growth, with billions of users and vast data centers making it a political issue in many states.
Anthropic, which is reportedly preparing for what could be the largest tech IPO ever, has tried to reassure investors and the public about its safety efforts. In a statement to WIRED, a company spokesperson pointed to Anthropic’s work on mechanistic interpretability and called for “a lawful, verifiable way to work together to pace how we release powerful models.” OpenAI declined to comment.
According to a Wall Street Journal report, Coxon announced his resignation from Anthropic and the AI industry on September 9, 2026, citing fears of "out-of-control" artificial intelligence and the risks of self-improving AI systems. He left about two months before his equity vested, a move he described as a deliberate sacrifice for principle, as reported in a Deadline article.
Coxon worked in pretraining research at Anthropic and was previously affiliated with OpenAI. His resignation, which occurred around September 8-9, 2026 after nearly three years at Anthropic, is seen as part of a broader rift within AI labs over the race to develop self-improving systems.
Coxon’s main warning centers on recursive self-improvement-using AI to build even more advanced AI. He wants Anthropic and OpenAI to coordinate and pause this practice, at least for now. He also argues that only international agreements, especially between the US and China, can prevent catastrophic misuse, such as AI-enabled biological threats or cyberattacks.
Coxon does not ignore the possible benefits. He notes that AI could lead to breakthroughs in medicine and science, including cures for diseases like cancer. But he insists that without enforceable safety rules, the risks outweigh the rewards. “If nothing is done to slow their race for dominance,” he warns, “both companies could cut corners in the future.”
The debate over AI safety is now playing out in boardrooms and government offices, with new regulatory proposals gaining ground. The Hugging Face breach is just the latest in a series of incidents forcing companies to confront the limits of their own technology. As previously reported, OpenAI’s leadership has also warned about the security risks of advanced models, urging businesses to strengthen their defenses before attackers exploit new vulnerabilities.
Coxon’s immediate proposal is a binding agreement among leading Western AI labs to avoid recursive self-improvement for now. But he sees this as only a temporary fix. In the long run, he says, governments will need to treat AI computing resources like dangerous materials, with strict oversight and international cooperation.
The industry is now in a standoff between safety and speed. Coxon’s resignation is more than a personal protest-it’s a warning to regulators, investors, and creators who rely on these systems. The next two years will be decisive. If industry leaders fail to coordinate and governments hesitate, the consequences could be permanent. For those building on AI-driven workflows, the message is clear: the rules are changing, and ignoring the risks could have lasting costs.
Anthropic was founded in 2021 by former OpenAI researchers and has quickly become one of the most heavily funded AI startups. The company has raised over $7 billion from investors including Google, Amazon, and Salesforce. Its Claude models now power applications used by millions. Anthropic’s valuation passed $15 billion in 2026 as it expanded cloud partnerships and prepared for a major IPO, making it a central player in the global race to commercialize safe, scalable artificial intelligence.
