UN’s AI experts warn against ‘apocalyptic’ rhetoric on risks

On Monday, several experts from the UN’s artificial intelligence committee voiced concerns over the alarmist language often used by industry professionals discussing AI risks. Joelle Barral, a Google DeepMind executive and member of the International Scientific Panel on AI, cautioned that investing in fear does little to advance the field.

Jacob Coxon, a former Anthropic employee who previously worked at OpenAI, had warned in early September that AI developers believed the technology could wipe out humanity by the end of the decade. Another research engineer who left Google DeepMind in July echoed this sentiment, suggesting it might already be too late to prevent humanity’s eradication by AI.

Nobel Peace Prize laureate Maria Ressa, an investigative reporter, stressed the need to move beyond such apocalyptic rhetoric, emphasizing that the reality of AI risks is far more nuanced and sophisticated. The panel, established in 2025, aims to deepen scientific understanding of AI and advise UN deliberations. Yoshua Bengio, a renowned AI researcher and co-chair of the panel, stressed the importance of differentiating between what scientists know about AI risks and what can only be plausibly extrapolated, highlighting the ongoing disagreement among experts.

He cautioned that excessive shouting can lead to others shouting back, and instead called for a rational global discussion that acknowledges uncertainty and the severity of potential risks.