What’s behind the AI industry’s latest warnings of doom?

The most recent spike in public alarm over AI safety followed a resignation and a high-profile social media amplification. Jacob Coxon left Anthropic, saying he was concerned the major AI firms were taking risks with human safety; an Anthropic alignment team member then shared Coxon’s words and added their own dire assessment, including a greater-than-10% personal estimate that AI could extinguish humanity within ten years. Those statements reverberated through the tech press and social platforms, prompting renewed debate about how seriously to take such doomsday predictions. On TechCrunch’s Equity podcast, hosts parsed motives and meaning behind the warnings. Some panelists expressed skepticism about broad claims framed as if the AI community were a single unified voice, and questioned the value of throwing around precise-sounding probabilities without clear methodology. Others noted the significance of someone leaving their job over safety fears, arguing that a resignation is a stronger signal than abstract rhetoric when assessing how worried insiders really are. Panelists also explored alternative explanations for the public alarm. One line of analysis suggested that vivid warnings can double as demonstrations of capability—if models are advanced enough to generate worrying failure modes, that in itself signals technical progress. There was also discussion of incentives: as companies like Anthropic and OpenAI iterate new models and prepare for potential public listings, stark safety claims could affect public perception, recruitment, or investor reactions. One host even raised the practical question of whether such assessments will find their way into legal documents like S-1 filings for IPOs. Listeners were reminded that the debate comes amid an uptick in reporting on model capabilities and security incidents — for example, a recent Hugging Face breach related to an OpenAI internal model and the rollout of new systems such as Anthropic’s releases and OpenAI’s Astra — which adds urgency to conversations about oversight and control. The Equity discussion ended with no easy resolution: participants acknowledged that some industry voices appear genuinely alarmed while others may have mixed motives, leaving open how companies will balance transparency, safety engineering and commercial pressures as AI systems continue to evolve.