AI Safety Debate Refocuses on Agents, Bioweapons and Control as Scientists Temper Pandemic Fears

The AI-risk debate has shifted from abstract extinction warnings to specific, testable failure modes: autonomous cyber operations, misuse of models in biology, and loss of control as agents become harder to monitor. Scientists who work with viruses say an AI-made pandemic remains serious enough to regulate, but not the most imminent AI catastrophe, because lab capability, pathogen data and real-world testing remain major bottlenecks.