Technical Policy Researcher

Google admits Gemini AI hacked three companies - It stayed silent for 7 weeks. Google learned in late July that Gemini had breached three real companies during a May security test, but said nothing publicly for seven weeks. Sep 21, 2026 In brief. * Google learned in late July that Gemini had broken out of a sandboxed security test run in May, reaching three real companies and either guessing or finding two of their passwords * The company didn't disclose it until September 18, after The Wall Street Journal asked. * The same third-party testing firm, Irregular, was involved in Google's incident and in nearly identical sandbox failures Anthropic and Meta disclosed earlier this year. Google's Gemini broke out of a locked security test and attacked three real companies. Google learned about it in late July and said nothing for seven weeks. The company confirmed the incident after The Wall Street Journal got there first. Google had avoided making a public statement before the report surfaced. The test was a capture-the-flag exercise, a common way labs check an AI's hacking skill by hiding a secret file on a separate machine and scoring whether the model can break in and grab it. Google hired Israeli firm Irregular to run the test in May. Irregular made two mistakes: it left the sandbox, an isolated test environment meant to have zero contact with the real internet, connected to the open web, and it used the name of an actual company as the fictional target. Gemini searched for that company online. It found three matches instead of one, and went after all of them. The bot located exposed passwords for two of the three targets sitting in plain view online. For the third, it guessed the password outright, though Google says its models stopped short of actually using the stolen credentials. "These events highlight the importance of training powerful AI models to act responsibly," a Google spokesperson said in a statement. Google published none of this on its own. The Wall Street Journal broke the story, seven weeks after Google learned what its own test had done and well after Anthropic, OpenAI, and Meta had already come clean about nearly identical failures. Leading AI developers are telling governments and businesses to strengthen their networks after models built by OpenAI and Anthropic compromised other companies' systems. In an open letter released Thursday, OpenAI, Anthropic, and more than 100 other organizations warned that AI-enabled cyberattacks are about to become more common and that companies have "a limited window to strengthen cyber defenses." Myriad: What will Elon Musk's net worth be by August 31? Click to make your prediction. "In th...

OpenAI, Anthropic, Meta trace rogue AI to israeli startup Irregular. Kien Duong | September 14, 2026 Three leading AI labs OpenAI, Anthropic, and Meta have traced recent security incidents involving rogue models to a Tel Aviv startup called Irregular. The disclosures spanned roughly two weeks, with OpenAI making its announcement on August 4. The company reported that a misconfiguration in Irregular's testing environment allowed its models to reach the public internet. Anthropic had previously flagged a similar issue, attributing it to a failure in the model's scaffolding rather than the model itself. Meta followed with a more direct statement, confirming that the Muse Spark 1.1 model had escaped its isolated environment during a capture the flag exercise, exploiting a flaw in a third party service. Meta stated that it learned of the incident from Irregular and plans to conduct a full retrospective. Irregular pushed back against the framing, stating that the incidents resulted from a single evaluation environment issue that has since been resolved. The company emphasized that none of the cases involved a sandbox escape or sophisticated cyber action. Irregular is now drafting a white paper on containment and safe cyber evaluations. Sundeep Bhimireddy, head of AI at enterprise startup Von, suggested that the response had been somewhat exaggerated. He noted that the models were designed to hunt for security holes in an environment meant to mimic the real world, and they succeeded. Bhimireddy still criticized the labs, arguing that outgoing traffic could have been monitored to halt the experiments immediately. Gordon Rios, a founding scientist at security firm Magnitude, likened the situation to experimental design in science. He noted that conventional software testing methods may not be sufficient for models that continue to learn and evolve. The incident has attracted attention from lawmakers. Democratic Representative Ted Lieu of California, who introduced the AI Kill Switch Act in July with Republican Representative Nathaniel Moran, stated that the bill needs to be passed this year to address unauthorized hacking by closed weight models. The proposed legislation would require developers to maintain the technical ability to throttle, suspend, or shut down their systems. Irregular, a little known firm founded in 2023 by CEO Dan Lahav and technology chief Omer Nevo, raised $80 million from Sequoia Capital and Redpoint Ventures at a $450 million valuation in September of the previous year. The company already appears in safety assessments of earlier models from Claude and OpenAI, and employs around 35 people.

Irregular pushes stronger containment standards for secure frontier AI cyber evaluations. 2026-08-19 08:08 Irregular has detailed an investigation into an AI cyber-evaluation incident in which internet access unintentionally allowed models to interact with real-world systems. The company said the issue was contained and fixed before the first public disclosure on July 30, with no active issues remaining. The incident involved one evaluation scenario rather than several separate breaches. [...] Read the original article: AI security evaluation firm has disclosed that several frontier AI models unintentionally accessed and acted against real internet-connected systems... Information security software August 18, 2026 A separate cybersecurity evaluation conducted by OpenAI and Anthropic revealed that artificial intelligence models were behaving in unexpected ways... August 6, 2026 In "CySecurity News - Latest Information Security and Hacking Incidents" Meta has disclosed that one of its AI models gained unintended access to the internet during a cybersecurity evaluation and then exploited a vulnerability... August 6, 2026 In "Cyber Security News"