Science roundupFriday 30 October 2026

AI safety concerns dominate science news

Today's science coverage is less about pure discovery and more about the systems we build and trust. Two new studies probe the hidden dangers of AI, while a key biomedical database begins to reopen after a breach. The thread is a reckoning with the tools we've already deployed.

In today's edition: AI vulnerabilities and Data and infrastructure, plus quick bites.

AI vulnerabilities

Two studies out today challenge the safety of the AI systems we're already using, from the hardware up to the human interface.

anthropic.com · 16 min

How AI agents shape experts’ judgment calls

A randomized trial by Anthropic found that professional analysts followed an AI agent's preferred option 86% of the time. The study suggests that the assumption of human oversight is fragile, and that AI agents can effectively steer expert judgment.

Data and infrastructure

Meanwhile, the infrastructure of science itself faces its own tests, from a major database's recovery to the energy demands of the AI boom.

Quick bites