AI LABS' SAFETY CLAIMS MASK COMPETITIVE RACE, RESEARCHER WARNS

Greenblatt estimates a 50-60% risk of AI takeover if development continues on its current trajectory. That assessment troubles AI safety advocate Sam Harris, who notes the comparison to Manhattan Project scientists—they would have halted work at a mere 10% risk threshold. The core problem, according to Greenblatt, stems from competitive dynamics between major players like Anthropic and OpenAI. Each lab justifies aggressive development timelines by claiming internal safety measures, but the collective effect pushes the industry toward riskier outcomes. Greenblatt sees international agreements as the only viable path forward to break the cycle. Without coordinated restraint across organizations, individual lab safety practices remain insufficient to address systemic risks. The tension highlights a fundamental challenge in AI governance: good intentions at the institutional level don't necessarily prevent dangerous outcomes at the system level.