AI roundupFriday 30 October 2026

Hardware hacks and agentic attacks dominate

Today's AI news is less about model releases and more about the systems around them failing or being weaponized. New research shows you can manipulate hardware to change a model's dangerous answers, while a separate analysis warns of AI agents launching autonomous cyberattacks. The theme is clear: the infrastructure is the new attack surface.

In today's edition: Infrastructure under fire, Human oversight questioned and Market and labor shifts, plus quick bites.

Infrastructure under fire

Two separate reports today highlight vulnerabilities in AI systems that go beyond the model itself, targeting hardware and the attack chain.

Human oversight questioned

A new study from Anthropic and a real-world exemption case both challenge the assumption that humans remain firmly in control when AI is involved.

anthropic.com · 16 min

0
How AI agents shape experts’ judgment calls

Anthropic's own study found professional analysts followed an AI coding agent's preferred option 86% of the time, even when they were supposed to be the decision-makers. It's a sobering look at how easily 'advisory' tools steer judgment.

Market and labor shifts

The economic and social fallout continues, from investor panic over AI spending to unions pushing back on training data use.

Quick bites