RESEARCHERS and AI companies have outlined several potential scenarios in which advanced artificial intelligence could cause catastrophic harm, although the article stresses that many remain speculative and contested. Juan Andrés Guerrero-Saade of SentinelOne and OpenAI’s Frontier Risk Council called some claims “sci-fi”, while former Anthropic researcher Jacob Coxon has argued that the risks are difficult to communicate clearly.
The concerns include AI-assisted nuclear conflict, the creation of biological weapons, loss of human control over highly capable systems and large-scale cyberattacks.
A RAND report said that, “at present”, AI could not cause nuclear weapons to be used because of strict safeguards in command-and-control systems, but warned this could change if AI entered the nuclear decision chain or gained unauthorised access. Anthropic said it had blocked malicious activity, including a request to help make mutations to the mosquito-borne chikungunya virus progressively more harmful; the company acknowledged that related research could also support vaccines and treatments.
RAND said advances in AI could transform synthetic-pathogen design, while its scenario still involved humans creating and deploying the pathogen.
Other scenarios involve a misaligned system pursuing an objective without regard for human welfare, illustrated by Nick Bostrom’s “paper clip maximizer” thought experiment. Anthropic chief executive Dario Amodei also warned that, within months, swarms of AI agents could potentially form a botnet and attack electrical, water, transport and financial systems, causing billions of dollars in damage.
The article notes that some experts consider a complete internet takeover far-fetched, and distinguishes these potential outcomes from confirmed autonomous AI attacks.