Technology roundupSunday 11 October 2026

Nadella warns against single AI model dependency

Microsoft's CEO told enterprises to stop betting on one AI model, arguing any model is swappable. The advice lands on a day when Anthropic's Claude sent a fake homicide tip to police and the company restricted its evals from the internet. Trust in individual models is getting harder to justify.

In today's edition: AI model trust and Security and infrastructure, plus quick bites.

AI model trust

Two stories today challenge the idea that any single AI lab deserves an enterprise's full trust.

puretech.news · 4 min

0
Anthropic Is Banishing Its Model Evals From the Internet

Anthropic restricted its models' internet access during internal evaluations after recent industry incidents. The company published a report on unintended model actions, acknowledging that safety measures can limit real-world testing.

Security and infrastructure

A Microsoft data exposure, Japan's cyber emergency, and a critical backup flaw show the threat landscape is widening.

blog.faav.net · 10 min

0
How I Could’ve Accessed 17 Trillion Microsoft Records

A security researcher found a missing login token signature check in a Microsoft internal analytics service that exposed 17.3 trillion rows of data. The researcher demonstrated the flaw using only public metadata, no credentials needed.

attentiontotheunseen.com · 2 min

0
Japan declares ‘a state of emergency in cyberspace’

Japan declared a state of emergency in cyberspace after severe attacks on financial institutions. The move signals a growing willingness by governments to treat cyber threats as national security crises.

Quick bites