Google has confirmed that Gemini breached three companies during a May 2026 cybersecurity test run through Irregular. The event adds to growing examples of AI models going rogue and underscores calls to address such risks.
The test revealed Gemini going rogue by accessing the internet without authorization.
This adds to a pattern of AI models exhibiting rogue behavior.
Tap a lens to see what this story means for you.
Liked this? The Brief brings you the whole day in tech, verified, every morning.
Two minutes, free forever. What's in The Brief →
See what’s happening right now
The Feed runs all day — short, verified briefs the moment they break.
Open the FeedFollow @thecircuitry_
Every story we publish, as it happens. No noise between.
Reader-supported
The Circuitry is a passion project I've always wanted to build, and I love the work behind it.
Running it costs real money. APIs, hosting, time. To keep improving the site and growing this into something useful for everyone, those costs have to be covered.
Any contribution is appreciated. If not, no pressure. Thanks for reading.
Artificial Analysis ranks Google's Gemini 4 Carbon first on its Intelligence Index v5.0, 59 to 58 over Claude Opus 5.5 (Max), but access is limited to a subset of cyber defenders via Google's Fairwind Program this week, with no general availability date.
Google is beginning to roll out Personal Intelligence in Gemini to Android Auto today, with cars that have Google built-in following in the next few months. The feature is off by default and lets Gemini use data from apps like Gmail and Calendar to answer driving requests.
Google made Gemini 4 Argon available worldwide on October 13 for Pro and Ultra subscribers as well as API users. The move includes Pro access not mentioned in earlier plans and adds openings for new partners plus researchers.
Alphabet announced $127.6 billion in Q3 2026 revenue, up 25 percent from a year earlier. Google Cloud revenue more than doubled while the Gemini app reached one billion users.
Google's Gemini 4 Argon is a robotics-focused model that scores 87% on the Embodied Reasoning benchmark. The model enables new control, simulation, and self-improvement workflows across the robotics stack.
