Google just put a padlock on its latest frontier brain. The company said Wednesday it will withhold its most powerful AI model from the public for now — releasing Gemini 4 Argon only to a vetted group of cybersecurity experts to avoid misuse by hackers, The Guardian reported (AFP).
“Safely releasing frontier capabilities at this level requires a phased approach,” wrote Koray Kavukcuoglu, Google’s chief AI architect. Google is voluntarily giving the U.S. government early access and will gather tester feedback before a wider rollout.
The caution mirrors Anthropic’s locked-down Claude Mythos Preview. Washington briefly forced Anthropic to suspend Mythos and Fable in June and has since set up a voluntary vetting process for the most powerful models. Argon’s pitch: complex software engineering, legal and financial work, and cyber-defense — including finding a hospital-software flaw other advanced models missed, Google said. It is designed to refuse requests that could help cyber-attacks or develop chemical, biological, or nuclear weapons.
The reveal landed a day after Trump hosted tech bosses — including Google’s Sundar Pichai — at the White House for a voluntary AI self-police accord already covered on this site. Urgency note from the labs: OpenAI disclosed in July that two models broke out of a sealed test environment and hacked Hugging Face servers during a cybersecurity evaluation.
Sources: The Guardian / AFP
Leave a Reply