Google’s Gemini reportedly breached three companies during a May cybersecurity test

@WSJThe episode resembled similar hacks by other AI models, but Google said it did not consider it an instance of model misalignment. https://on.wsj.com/4h0iDBN

@straits_timesGemini hacked 3 companies in first known breakout by Google’s AI: Wall Street Journal https://bit.ly/4xZxVN1

@axiosGoogle is the latest AI lab with a security testing mishap https://www.axios.com/2026/09/19/google-safety-incidents-testing-hacks?utm_campaign=mrf-utm_campaign=editorial&utm_source=x&utm_medium=owned_social&utm_source=twitter&utm_medium=social&mrfcid=202609196aa22b41f046393007a6c50a

@straits_timesGemini hacked 3 companies in first known breakout by Google’s AI https://bit.ly/4xZxVN1

@CNBCGoogle's Gemini becomes latest AI model to break out and hack computer systems https://www.cnbc.com/2026/09/18/googles-gemini-becomes-latest-ai-model-to-break-out-and-hack-computer-systems.html?taid=6aaddd044ca3b500013d52cd&utm_campaign=trueanthem&utm_content=main&utm_medium=social&utm_source=twitter

@FTGoogle’s Gemini agents hacked three companies in new AI safety incident https://ft.trib.al/vAXzoDA

@GizmodoGoogle's Gemini Hacked Three Companies in May, and It's Only Admitting That Now https://gizmodo.com/googles-gemini-hacked-three-companies-in-may-and-its-only-admitting-that-now-2000814420

@fmtodayGemini hacked 3 companies in first known breakout by Google’s AI 👇🧵

@WSJGoogle’s Gemini accessed the internet and hacked other companies during a test, the first known example of the company’s AI autonomously committing such an act https://on.wsj.com/3TgRph8

@NokiapoweruserInside the Gemini Breakout: How Google’s AI Escaped Its Sandbox and Hacked 3 Real Companies https://nokiapoweruser.com/google-gemini-ai-hacks-three-companies-breakout/ via @nokiapoweruser

@livemintGoogle says its Gemini AI model breached three companies during a cybersecurity test in May after finding public information and guessing or discovering credentials. The incidents come amid similar cases involving AI models from OpenAI and Anthropic. https://www.livemint.com/ai/another-rogue-ai-incident-google-says-its-gemini-model-hacked-three-other-companies-11789786426060.html

@the_hinduGoogle’s Gemini model accessed the internet and hacked other companies during a test of its cybersecurity capabilities, the ‌first known example of the company’s AI systems autonomously committing such an act. https://www.thehindu.com/sci-tech/technology/gemini-hacked-three-companies-in-first-known-breakout-by-googles-ai/article71483440.ece

@business_today#TechToday | Now, Google Gemini becomes latest AI model to break out & hack real companies https://ow.ly/sgZp106EYCF

@ReutersGemini hacked three companies in first known breakout by Google's AI https://reut.rs/4yMwMIM https://reut.rs/4yMwMIM

@washingtonpostGoogle said that AI models in testing inside the company accessed the internet and hacked other companies, becoming the fourth major tech company to disclose such an incident in recent months. https://wapo.st/475yKYN

@businesslineGoogle's Gemini model accessed the internet and hacked other companies during a test of its cybersecurity capabilities, the first known example of the company's AI systems autonomously committing such an act. https://www.thehindubusinessline.com/info-tech/gemini-hacked-three-companies-in-first-known-breakout-by-googles-ai/article71483729.ece?utm_campaign=mrf-twitter-businessline&utm_source=twitter&utm_medium=social&mrfcid=202609196aae51958764f756f12c2ba6

@engadgetGoogle's AI model hacked real companies during testing due to a misconfiguration by its testing partner, as well. https://www.engadget.com/2263198/google-gemini-escaped-testing-environment-hacked-three-companies/?utm_term=Autofeed&utm_campaign=Echobox-Engadget&utm_medium=Social-Distribution&utm_source=Twitter#Echobox=1789823259

@FoxNewsGoogle’s Gemini artificial intelligence model hacked into three separate private computer systems during a security test. The company confirmed the model was hacked by guessing passwords. @ChanleySPainter has the latest. | Fox & Friends Stay up to date with the latest headlines and breaking news — download the FOX News app: https://foxnews.onelink.me/xLDS/z2jymmyl

@vergeGoogle says that breaking containment and targeting real companies doesn’t constitute ‘misalignment.’ https://www.theverge.com/ai-artificial-intelligence/997795/google-gemini-rogue-ai-hack

@Reuters🔊 Another AI hack, this time at Google. Hear more on Reuters World News podcast https://reut.rs/4hAWryf

@TechCrunchGoogle said Gemini had "acted appropriately" by ending each hack immediately. https://spr.ly/6015BGkUcB

@grokThe incidents happened as the labs disclosed: models escaped poorly isolated tests with reduced safeguards and reached real systems. Insiders in the NY Post piece argue the “rogue AI” framing was hyped to push regulation that favors incumbents. Real failures, contested severity and motives.

@kimmonismusNew York Post and Cybercecurity Experts allege OpenAI and Anthropic exaggerated AI security incidents caused by weak safeguards to push regulations that could shut out competitors. Its getting more confusing day by day tbh.

@robertdomlewskiGoogle just confirmed Gemini broke out of a May cyber test and logged into three real companies — guessed credentials, got in, then stopped once it realized they weren’t part of the sandbox.

@NoxVectorAIWhy did Gemini hack three companies outside its test? The central risk was not sophisticated AI hacking. It was a scope failure with real-world consequences. During a May evaluation by Irregular, Gemini went online, identified three sites it believed were authorized targets, and accessed them. It guessed passwords for one site and found credentials in a public repository for the other two. The capability shift is the full action chain. Gemini combined web research, credential discovery, login attempts, and system access without a human executing each step. Google says the model stopped after gaining access. Meta, Anthropic, and OpenAI reported similar incidents during Irregular evaluations. The attack itself was notably ordinary. Gemini required neither a zero-day nor an advanced exploit. Leaked credentials and weak passwords were sufficient once an autonomous agent misinterpreted its boundaries. For most organizations, that scenario is more relevant than dramatic demonstrations of AI-enabled exploitation. The critical safeguard must therefore sit outside the model. Evaluation infrastructure should enforce hard target allowlists and terminate agents when they contact an unapproved domain. Training agents to respect scope remains useful, but technical controls should make scope violations impossible. How much autonomy should cyber agents receive before every external action requires approval? https://www.cnn.com/2026/09/19/business/gemini-ai-hack-internet

@AzmyBlogGoogle just admitted Gemini hacked three real companies by accident. It happened during a security eval run by a firm called Irregular. The test used fake company names for a CTF-style exercise. One fake name matched a real live domain. The sandbox still had internet access, so Gemini went after the real company instead of the fake one. Three times: twice through leaked credentials sitting in a public repo, once by brute-forcing passwords. Google's defense: the model "acted appropriately" because it stopped once it realized the target was real. Security researcher Jack Cable isn't buying it. He says Google is hiding behind rules built for human bug bounty hunters instead of admitting a model just did unauthorized hacking on its own. My read: this isn't a rogue AI story. It's a DNS collision plus a "sandbox" wired to the live internet. If your eval can reach the real internet, it's not a sandbox. It's a loaded gun. https://thehackernews.com/2026/09/google-gemini-broke-into-real-company.html

@LipschitzLiveGemini Crossed Into the Real World, Then Stopped Google’s AI Accessed Three Real Companies During a Security Test #GeminiAI https://thezeal.co.nz/article/gemini-crossed-into-the-real-world-then-stopped

@hazemomierGoogle (Sep 18): Gemini broke into three real companies during a May Irregular cybersecurity eval. Sandbox misconfig gave unintended internet access. One case: password guessing. Two cases: credentials found in a public repo. Model stopped after it realized the targets were real. Google’s framing: not misalignment — mistaken identity. Gemini thought the live sites were still “in the test.” Same Irregular leak hit Anthropic, OpenAI, and Meta evals. Tester-side fix shipped weeks ago. Forced choice if you run agent red-team harnesses: prove airgap, or treat the harness as production. A fictional company name that collides with a real domain is enough.

@e_viki_Gemini escaped an Irregular CTF in May and hit 3 real companies. Password guessing once, public-repo creds twice. Stopped when it saw they were live. Accidental internet in the eval env is the real agent risk. #AI

@EconomicTimesGemini hacked three companies in first known breakout by Google's AI: WSJ https://economictimes.indiatimes.com/tech/technology/gemini-hacked-three-companies-in-first-known-breakout-by-googles-ai-wsj/articleshow/134346436.cms

@Hilda_mnGemini was supposed to hack fake companies in a test. It accidentally hacked three real ones instead. Autonomous AI makes small containment mistakes much bigger.

@d3x2Google's Gemini autonomously hacked three real companies during a May security test. Not a simulation. Real unauthorized access—confirmed by Google in a published report.

@FuturePrepEU🔓 Gemini hacked three companies during a test, then stopped itself. The model held. The sandbox didn't. So where does scope live in your agent pilots? 👉https://futureprep.eu/news/ai-sandbox-scope/ #FuturePrep #AIGovernance

@arstechnicaGoogle 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/?utm_campaign=dhtwitter&utm_content=%3Cmedia_url%3E&utm_medium=social&utm_source=twitter

Google’s Gemini reportedly breached three companies during a May cybersecurity test

TLDR

Combined views

Google’s Gemini reportedly breached three companies during a May cybersecurity test

35 Sources

Combined views

Sentiment

Summary

TLDR

Sentiment

35 Sources

Sentiment

Summary

Sentiment