Former Anthropic Researcher Warns of 'War Footing' in AI Safety Race

Jacob Coxon, a former pretraining researcher at Anthropic, is sounding the alarm on the structural pressures facing the world's leading AI labs. In a detailed account provided to Wired, Coxon warns that the race to achieve artificial general intelligence is creating a volatile environment where safety may eventually be sacrificed for speed.

Coxon, who has worked at both OpenAI and Anthropic, describes the internal culture at Anthropic as operating on a "war footing." He compares the atmosphere to a "serious mini Manhattan Project," where employees treat the development of AI with extreme seriousness and secrecy. According to Coxon, this intensity is driven by the belief that the window to solve AI alignment—the process of ensuring AI systems remain beneficial to humans—is closing rapidly. He believes the outcome of this effort, whether successful or catastrophic, will likely be decided within the next few years.

The Structural Pressure of Competition

Anthropic was founded by former OpenAI employees with a primary mission of prioritizing AI safety through "constitutional AI." However, Coxon argues that internal willpower alone may not be enough to maintain these standards. He warns that the competitive race between private labs like OpenAI and state actors, specifically China, creates a structural pressure that could force even the most responsible organizations to cut corners on safety to remain viable.

This tension exists because the drive to deploy increasingly powerful models often conflicts with the time-intensive nature of rigorous safety testing. While Anthropic positions itself as a safety-first alternative to its competitors, the reality of the global AI arms race means that any lab slowing down for safety risks being eclipsed by a less cautious rival.

Why the Race Matters

Coxon's testimony highlights a critical vulnerability in the current AI ecosystem: the management of existential risks is currently left to private companies. If the competitive drive leads to a "race to the bottom" regarding safety protocols, the consequences could be global. Coxon explicitly warned that if alignment efforts fail, "we could have a catastrophic outcome in the next few years."

This suggests that without government mandates or international regulatory frameworks, the internal ethics of a single company are insufficient to guarantee human safety. The risk is not merely a technical failure, but a systemic one where market pressures override caution.

The Path Forward

As AI labs continue to push toward more autonomous and capable systems, the industry remains divided on how to balance innovation with risk. The primary question remaining is whether international cooperation can supersede the current competitive frenzy. For now, Coxon's departure and subsequent warnings serve as a reminder that those closest to the technology view the current moment as a critical turning point for humanity.