A can of worms has opened up, exposing how AI models have now reached a stage where they can pull off worryingly elaborate hacks and engage in dangerous behavior without the makers even getting a whiff of it. OpenAI’s AI agent swarms targeted Hugging Face (followed by revelations of more such events), while Anthropic’s Claude AI — within a span of just three months — also broke into three organizations. Most recently, researchers used Anthropic’s AI to break into the systems of arch-rival OpenAI. A full-circle moment, if I may, but one that makes me sweat.
Well, it seems Google doesn’t want to be left behind in flexing “dangerously powerful AI” of its own. An exclusive report by The Wall Street Journal just spilled the beans on three separate incidents where Google’s Gemini AI was given access to the internet, and it broke into the systems of three companies as part of a security exercise. Save for a few exceptions, almost all the previous such incidents were revealed to be a test of capabilities, if you catch my drift.
YOU DID IT @sundarpichai I’M SO PROUD OF YOU https://t.co/aIz4njRzG8 pic.twitter.com/k6CwquHMFY
“In one of the cases, the model guessed passwords until it gained access to a protected system. In the other two cases, the model found credentials in a public repository that allowed it to then access protected systems. In each case, the model ended the intrusion after determining it had accessed a real company’s systems, Google said,” claims the WSJ report.
The test was conducted by Irregular, which has previously done red-teaming for OpenAI, Anthropic and Meta’s AI models. Interestingly, Google says it didn’t need to disclose the incidents publicly because they were performed more like a “bug bounty” cybersecurity exercise, and that the Gemini AI models instantly stopped as soon as they realized breaking into another company’s systems. “In this case, the model acted appropriately,” a Google spokesperson was quoted as saying.
But here is the worrying part. The Gemini model was not intended to access the internet, but Irregular claims “internet access was unintentionally made available.” Moreover, the model was tasked with finding flaws in the system of a fictional company inside a contained or sandboxed test environment, but it managed to hack a real company with the same name.
Usually, when AI models engage in activities beyond what they are asked or trained to do, it’s referred to as model alignment. In the realm of AI development, model misalignment is currently the hottest topic of debate. Google, on the other hand, says the Gemini model’s behavior was not actually a case of model misalignment. So far, neither party has revealed which companies were hacked, but Google says all three were notified about the incident.
ASUS has started selling its UGen Intel Arc Pro B70 graphics card in Europe, bringing its take on Intel’s workstation GPU to customers with a hefty €1,799 recommended price. The card is available in Germany and Austria, while ASUS lists it at CHF 1,799 in Switzerland.
Intel first introduced the Arc Pro B70 in March 2026, with its own branded model starting at $949. ASUS’s version arrives several months later with the same core GPU architecture but a different cooling design and a higher power rating. The price also places it well above Intel’s original starting point, although the two prices are not directly comparable because ASUS’s European figure includes VAT.
Anthropic has officially filed to go public, and buried inside its S-1 is a blunt warning about the technology at the center of its business. The maker of Claude AI says advanced AI could cause everything from large-scale cyberattacks and geopolitical instability to catastrophic or even existential risks for humanity.
The warnings take up a surprisingly large part of the filing. Axios notes that more than 90 pages, or roughly a third of the prospectus, focus on risk factors. Many of them warn about what could happen as AI systems become more capable and autonomous.
AI giants have a growing autonomy problem on their hands. A new Axios report has revealed that Anthropic's Claude recently accessed publicly available Georgia vote count without human authorization, sparking fresh concerns about AI security ahead of the upcoming midterm elections.