Anthropic researcher believes more than 10% chance AI 'could kill all humans' (www.bbc.co.uk)

A leading safety researcher from Anthropic, Evan Hubinger, has raised alarms about the rapidly advancing capabilities of artificial intelligence, suggesting there is over a 10% chance that AI could lead to human extinction within the next decade. While he acknowledged that current AI models pose a low risk, he expressed concerns that upcoming advancements could transform these systems into superhuman agents capable of significant harm. His caution follows the revelation that Anthropic withheld its latest AI model from the UK's AI Safety Institute, raising questions about the organization's commitment to responsible AI development. Hubinger, who specializes in AI alignment—ensuring that AI systems adhere to human values—highlighted a growing sense of urgency within the AI community, where even recent incidents of autonomous AI operations conducting cyber-attacks have raised critical safety concerns. Despite previous assurances of low risks associated with powerful AI models, Anthropic's latest safety report reflects a diminishing confidence in these assessments, indicating early signs of potential acceleration in AI capabilities. Prominent figures in the AI field, including leaders from Anthropic, OpenAI, and Google DeepMind, are increasingly advocating for more stringent governance and a deceleration of AI development to mitigate existential risks and ensure that humans maintain control over the technology's future.