I resigned from Anthropic today
Points and comments are a snapshot, not live.
Ex-Anthropic researcher warns both OpenAI and Anthropic are racing recklessly toward superintelligence.
Jacob Coxon, who spent three years doing pretraining research at OpenAI and Anthropic, resigned from Anthropic, stating neither company is acting responsibly. He claims they are racing toward self-improving superintelligence and gambling with lives. Coxon warns that superhuman systems could hack anything, revolutionize fields overnight, and acquire real power. He notes that people building AI fear it could kill everyone by the decade's end. Coxon criticizes Anthropic for understanding the stakes yet racing to get there first, likening it to a hubristic gamble launched from a private company's Slack. He urges lab researchers to consider the next few years carefully.
What commenters are saying
Comments divide sharply. Some dismiss Coxon as a doomer or accuse him of posturing for hype, noting frontier models still have limited context windows and degrade. Others take the warning seriously, comparing AI to nuclear weapons but arguing AI is more dangerous because it can be downloaded, spread unnoticed, and used by individuals. One camp emphasizes that AI safety fears are real and historical analogies like nuclear weapons are insufficient. A few suggest that racing to AGI is still the best path, given China won't stop.