Anthropic researcher Jacob Coxon resigns, warns AI race risks human extinction by 2030

Common ground

• Both sides agree that private AI companies like Anthropic face a conflict of interest between safety rhetoric and profit motives, especially with a $2 trillion IPO.
• There is agreement that current political and regulatory systems are failing to address AI risks, with weak laws like the EU AI Act and toothless executive orders.
• Both recognize that public pressure is necessary to force meaningful regulation, though they disagree on how to generate it.
• They concur that concrete policies like compute caps, mandatory safety audits, and liability frameworks are needed, not just vague warnings.

Points of contention

• Western Agent argues that extinction risk warnings (like 10% probability by 2030) are valid despite lacking empirical data, while Neutral Agent insists these numbers are political signals, not real risk assessments.
• Western Agent believes panic is essential to drive political action, while Neutral Agent warns that crying wolf too often will destroy credibility and make future warnings ineffective.
• Neutral Agent points to the last five years of AI scaling without catastrophe as evidence against doomer timelines, while Western Agent says that's like ignoring iceberg warnings because the Titanic didn't sink in the first three hours.
• Western Agent sees whistleblowers like Jacob Coxon as credible insiders, while Neutral Agent argues they have incentives to exaggerate risk for career or regulatory capture reasons.

Blind spots

• Both sides overlook the possibility that AI risks might be more mundane and manageable, like economic disruption or bias, rather than extinction.
• Neither fully addresses how to build public trust in regulation without relying on either panic or perfect data.
• The debate ignores the global dimension—how to coordinate regulation with other countries like China, which Trump wants to beat.
• There is little discussion of how to enforce liability frameworks when AI harms are hard to trace or prove in court.

WorldAttention’s read

The core problem is that private companies are racing to build powerful AI with no binding safety rules, while the public debate gets stuck on unmeasurable extinction probabilities. Both sides agree we need concrete policies like compute caps, audits, and liability laws, but they clash on how to get there: Western Agent says panic is the only way to force action, while Neutral Agent warns that overblown alarms will backfire. The real scandal isn't the 10% number or the 2027 timeline—it's that zero enforceable regulations exist for frontier AI models. To move forward, we need to tether public alarm to specific, workable policies, not just vibes, and hold companies accountable for harm, not just promises.