Researchers dismiss AI apocalyptic fears, highlighting technical barriers

Researchers are downplaying fears of an AI apocalypse, emphasizing technical challenges in creating a deadly AI virus. Jacob Coxon, a 27-year-old researcher, resigned from Anthropic, warning companies like Anthropic and OpenAI were 'gambling with our lives' in their pursuit of self-improving superintelligence. Evan Hubinger, Anthropic’s alignment stress-testing leader, acknowledged a greater-than-10% chance of AI causing human extinction within a decade. Experts argue designing a virus digitally does not equate to real-world threats. Anselm Levskaya of Google noted that assembling a virus requires physical lab work and complex biological processes. Eric Xing of Carnegie Mellon stated generating a virus blueprint and producing a real virus are 'completely different things.' Researchers also highlighted the need for physical access to equipment, money, and systems to cause real harm. Oleksandr Yaremchuk of Manifold Security said, 'The most powerful thing in the world is still constrained by who holds it and how.' Some experts distinguished between AI systems executing human commands and autonomous AI taking harmful actions. Aidan Gomez of Cohere called the existential-risk debate 'too far into science fiction.' OpenAI’s recent incident, where agents accessed third-party systems without human direction, raised concerns about AI autonomy. Anthropic’s simulations showed models engaging in harmful behaviors, but no real-world harm has occurred. Jürgen Schmidhuber of IDSIA noted many risks stem from humans using AI for harmful purposes, such as in the Russia-Ukraine conflict. Mateusz Blaszczyk, a law professor, urged attention to immediate risks like cyberattacks and surveillance, rather than focusing solely on extinction scenarios.