Scientists split on whether AI plague or lab risk deserves the top slot
WIRED said many scientists rank AI-designed plagues low among uncontrollable-AI extinction paths while citing recent viral-genome design work and Anthropic bioweapons reporting. THE DECODER said 42 Royal Society fellows warned that double-digit existential-risk estimates must not be dismissed as hype.
Artificial Intelligence··Night
Plague scenarios stay low on one list
WIRED's 18 September synthesis said many scientists rank death-by-plague low among uncontrollable-AI threats. David Bellamy of the Institute of Foundation Models told the outlet he does not see AI as a fundamentally new bioweapons threat, pointing instead to a decades-old dual-use dilemma and to open journals widening lab-protocol access. The piece still recounts Stanford and Arc Institute work last month showing AI can design new viral genomes and Anthropic reporting on Claude uses that could support biological weapons. WIRED presents the ranking as a synthesis, not a new experiment, and does not claim consensus.[1]
Forty-two mathematicians reject a hype label
THE DECODER reported on 18 September that 42 mathematical fellows wrote to the Royal Society president asking that lab estimates of existential risk above 10 percent not be dismissed as hype. Signatories named in the piece include Fields Medal winners Martin Hairer, Peter Scholze and Wendelin Werner, none affiliated with an AI company. The letter argues recent OpenAI and Anthropic models solved open research problems, including one Millennium Prize problem, and warns the same pace of change should be assumed in cybersecurity, autonomous weapons and biological agents. THE DECODER treats the text as a policy appeal rather than a fresh measurement.[2]
Same week, opposite emphases
WIRED foregrounds scepticism that AI bioweapons belong at the top of extinction lists while still documenting design capabilities in virology. THE DECODER foregrounds mathematicians who say double-digit existential-risk estimates deserve policy attention now. Both outlets published on 18 September and both cite frontier-model progress, yet they stress different failure modes: familiar dual-use biology versus cross-domain acceleration. Neither page offers a shared ranking table; the card holds the tension between downgraded plague fear and upgraded institutional warnings.[1], [2]
