Google rolls out its most powerful AI model yet but bars the public over fears hackers could use it to attack hospitals

From the Newshound archive · first published 1 October 2026 · last updated 2 October 2026.

Google's Gemini 4 Argon can autonomously identify, confirm and remediate critical software vulnerabilities, with access initially limited to trusted cybersecurity organisations through the Fairwind Program.

Google has released Gemini 4 Argon, its most advanced artificial intelligence model to date, limiting access to screened cybersecurity professionals amid fears the technology could be exploited to attack banks, hospitals and government systems. Early testers have already used Argon to identify a critical vulnerability in hospital software deployed worldwide. The model outperforms rivals from OpenAI and Anthropic on several key benchmarks, though it trails on some coding metrics.

Google is giving the Trump administration early access as part of a voluntary vetting process before any wider public release.

Argon can generate up to one million tokens in a single task — a dramatic leap from the previous 64,000 limit — and costs $2 per million input tokens.

Google scrapped its planned Gemini 3.5 Pro release entirely; the delay came after a major overhaul of its DeepMind lab and the departure of several Gemini team leaders.

The GuardianGoogle rolls out new Gemini AI model but restricts access over safety concerns The IndependentGoogle finally updates Gemini with new top model – but says it is so dangerous only select users can have it Barron'sGoogle Unveils New AI Model Gemini 4 Argon, Sending Alphabet Stock Higher - Barron's Gulf BusinessGoogle raises AI stakes with Gemini 4 Argon, its most powerful model yet - Gulf Business The Rundown AIGoogle unveils Gemini 4 Argon with strong scores and limited access - The Rundown AI

Read today's full feed at The Newshound →