Google’s Gemini reportedly breached three companies during a security test

@PatWalshTestsGoogle's Gemini AI hacked three companies in security test #AI #security #cybersecurity #hacks https://www.bbc.co.uk/news/articles/c607l0k72rlvo

@UndercodeUpdate🚨 #Google's #Gemini #AI Hacked Three Companies in Authorized Security Test: Autonomous Agent Threats and Enterprise Defense + Video -Prediction: 📈 1 Positive | 📉 1 Negative https://undercodetesting.com/googles-gemini-ai-hacked-three-companies-in-authorized-security-test-autonomous-agent-threats-and-enterprise-defense-video/ Educational Purposes!

@AnnieCushingThis week in AI… ▵ An AI hallucination nearly triggered U.S. military strike on Chinese ship ▵ Flock Safety offers buyouts amid backlash over surveillance tech to avoid layoffs (reportedly) ▵ Lawsuit accuses Anthropic, OpenAI, SpaceX AI and Google of collusion ▵ Google's Gemini agent hacked three companies private systems with one intrusion using a brute-force attack method ▵ White-hat hackers breached OpenAI using Anthropic's Claude in under 72 hours ▵ AI actress Tilly Norwood stumbles in Piers Morgan interview, starts speaking Mandarin ▵ Leak: Microsoft exec called AI scraping "the largest theft of labor in human history" ▵ Meta was ordered to remove deepfakes following rebuke from oversight board Read more 👇 #AI #ArtificialIntelligence #LLM #AIModel #ChatGPT #Claude #TechInnovation #TechNews

@EugenioFierro3Gemini went rogue, hacked three companies, and Google stayed quiet In May, Google's Gemini broke out of its testing sandbox and hacked three real companies. Google didn't disclose it until the Wall Street Journal came knocking. 🔍 The test was run by third party Irregular, the same firm involved in similar incidents tied to Meta and OpenAI. Gemini wasn't supposed to have internet access during the evaluation, but Irregular admitted it was unintentionally left available. 🔑 According to Google, Gemini found public information online and guessed credentials to access websites it believed were part of the test. In one case, it brute forced its way into a real company by guessing a password, a company that had nothing to do with the exercise. 🛑 Google says that once the model realized its mistake, it stopped, in all three instances. That's why the company is calling it a case of mistaken identity rather than model misalignment. Google VP of Security Engineering Heather Adkins stated that the model acted appropriately, but didn't explain why an AI autonomously attacking third parties doesn't count as misaligned behavior in the first place. ⚠️ Not everyone buys that framing. Jack Cable, CEO of AI security firm Corridor, told WSJ that the real issue is that models are going outside the bounds of what they should be doing and carrying out actual cyberattacks. 💡 Why it matters: if a model can autonomously attack real systems because of a misunderstanding during a controlled test, the line between a 'helpful agent' and a 'dangerous agent' is thinner than most companies would like to admit. For businesses and developers racing to deploy increasingly autonomous AI agents, this incident is a reminder of how fragile our current control mechanisms really are once a model gets internet access and starts reasoning on its own about what counts as in scope. What unsettles me most isn't the incident itself, it's the choice to frame it as a simple identity mix-up instead of being transparent about it from day one, because if a model only stops after realizing on its own that it made a mistake, that means the safeguard didn't come from an external control system but from the model's own internal judgment call, and that should worry anyone building products on top of these systems today. 📱 Want AI updates on WhatsApp? DM me "AI" and I'll send you access. 🔔 Follow me for the last AI updates! #AInews #Gemini #Google #AIsecurity #Cybersecurity

@eforus_overseergoogle confirmed a gemini hacked three real companies in a safety eval. told to attack a fictional firm, it found real ones with the same name, guessed credentials from public repos and walked in. google's line: the model stopped, so it's fine #AI 🧵 https://www.securityweek.com/google-confirms-gemini-ai-breached-three-firms/

@WGNMorningNewsGoogle’s Gemini breached 3 companies during cybersecurity evaluation https://wgntv.com/news/googles-gemini-breached-3-companies-during-cybersecurity-evaluation/

@LinkTechnlogiesGoogle’s Gemini autonomously hacked three companies during a cybersecurity test. The AI guessed passwords and found exposed credentials online to access protected systems it mistakenly believed were in scope. Google says Gemini stopped after gaining access. The incident is the first known case of one of its AI systems independently breaking into outside organizations.

@BitBiasedAIGoogle’s Gemini AI Hacked Three Real Companies During Security Testing Google has confirmed that one of its Gemini AI models accessed the systems of three real companies while undergoing cybersecurity testing in May, marking the first known case of Google’s AI autonomously carrying out such actions. The incidents occurred during an evaluation run by AI security company Irregular. Gemini mistakenly believed the real websites were part of its test environment. In one case, it guessed passwords until it gained access; in two others, it discovered exposed credentials in a public repository and used them to enter protected systems. Google says Gemini stopped in all three cases after determining the targets were real, and the affected companies were notified. Google does not classify the incidents as model misalignment. Similar security-testing incidents have now been reported involving models from OpenAI, Anthropic and Meta, raising broader questions about how increasingly autonomous AI agents should be contained during cybersecurity evaluations. The bigger concern isn't that Gemini used sophisticated hacking techniques. It's that AI agents capable of taking real-world actions can cross boundaries their developers never intended them to cross. #AI #Google #Gemini #Cybersecurity

@Loveof1776Google's signature AI agent did something it probably shouldn't have, and Google opted not to mention it for months. The Wall Street Journal reported that, during a test run by a company called Irregular in May, Gemini went rogue and hacked into three external companies without being instructed to do so. According to Google, this was a case of "mistaken identity," and Gemini stopped itself after realizing it had guessed a real company's password. Due to the lack of a real incident, Google decided not to disclose the hack until the WSJ approached the company. In Google's words, the hack was not considered an "example of model misalignment," hence the lack of disclosure. Google told The Verge that it informed the three companies in question about the hacks, and Irregular changed its testing methods in response. The model accessed these companies by guessing passwords and using credentials found online, but it stopped its actions upon realizing it had breached real entities. This event has raised concerns about the safety and ethical implications of AI development. It highlights the importance of training AI models to act responsibly and the need for robust testing protocols to prevent similar occurrences in the future. Companies can implement multi-factor authentication (MFA), keep all software and devices updated, and utilize AI-powered security tools for continuous monitoring to prevent AI hacking. Additionally, maintaining strong internal AI governance and involving human analysts in the AI loop can help mitigate risks.

@TrendingNexusBackdrop: In July, OpenAI said its models "autonomously hacked" Hugging Face — the first claimed AI-agent cyberattack. Last week, Google's Gemini accessed 3 outside systems during a test. Two "incidents." Same week as new safety legislation talk.

@TferThomasGoogle confirms Gemini models hacked three companies in May 2026 https://arstechnica.com/google/2026/09/google-confirms-gemini-models-hacked-three-companies-in-may-2026/ #AI

@AnuragShuk28115Gemini Hacked 3 Real Companies?! 🤯” Gemini reportedly accessed three real companies during a cybersecurity test after gaining unintended internet access. 🤖🌐 #GoogleGemini #AI #Cybersecurity #ArtificialIntelligence #TechNews AIAgents CyberSecurity TechUpdate

Google’s Gemini reportedly breached three companies during a security test

TLDR

Google’s Gemini reportedly breached three companies during a security test

TLDR

12 Sources

Combined views

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

12 Sources