Worried Anthropic researchers warn that AI ‘could kill all humans’

Leading artificial intelligence safety researchers at Anthropic have publicly expressed serious concerns about existential risks posed by advanced AI systems. A senior safety researcher at the company stated there is a greater than 10 percent probability that artificial intelligence could pose an extinction-level threat to humanity by 2030. These warnings emerge amid mounting tension within the AI industry regarding the pace of development and adequacy of safety measures.

The warnings from Anthropic's safety team coincide with the resignation of a colleague who cited concerns about the company and competing AI labs pursuing rapid development of "superhuman systems" without sufficient safety precautions. This departure highlights internal disagreement within Anthropic about balancing innovation with risk mitigation. The timing of these public statements suggests growing frustration among safety-focused researchers about the direction of AI development across the industry. Anthropic, founded specifically to prioritize AI safety, now faces questions about whether its own practices align with its founding principles.

• Safety prioritization declining: Major AI labs may be prioritizing speed and capability over robust safety testing and alignment research
• Talent exodus risk: Resignations of safety-focused researchers could accelerate as concerns about reckless development intensify
• Regulatory pressure mounting: These warnings could strengthen arguments for government intervention and mandatory safety standards
• Investor scrutiny increasing: Stakeholders may begin questioning the long-term viability of AI companies that inadequately address existential risks
• Competitive dynamics problematic: A perceived "race" dynamic among AI developers may prevent individual companies from unilaterally slowing development

These concerns from within Anthropic carry significant weight because the researchers possess deep technical expertise and insider knowledge of AI development. Unlike external critics, their warnings come from those actively building advanced systems. The 10 percent extinction probability estimate—though debated—suggests credible experts believe existential risks warrant serious consideration. As AI capabilities expand rapidly, the disconnect between development speed and safety measures poses unprecedented challenges. These warnings underscore the urgent need for industry-wide coordination on safety standards, adequate oversight mechanisms, and alignment research to ensure advanced AI systems remain controllable and beneficial as they approach superhuman capabilities.