Google says its AI model gained unauthorized access to three outside systems

NBC News ^ | 09/21/2026 | David Ingram and Jared Perlo

[SeekAndFind]

Google on Friday disclosed the first known instance of its artificial intelligence software, Gemini, carrying out an undirected computer hack, weeks after similar disclosures by AI firms Anthropic and OpenAI raised security alarms about AI models going beyond the instructions of their human creators.

Google said in a statement that in May its AI model gained unauthorized access to three outside systems during a test by either guessing login information or using login credentials it found in a public repository.

Heather Adkins, a Google vice president for security engineering, said in the statement that the AI model thought that the outside computer systems “were part of the test,” but she said in all three instances, the model stopped before doing anything further with its access.

“In a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test,” she said.

Google said it did not consider the unauthorized logins to rise to the level of misalignment, the AI industry term for software going rogue or not following instructions. Instead, the company said, the intrusions resulted from mistaken identity, where Gemini thought it was operating within a test but was actually connected to the real internet. Google said the model corrected itself and the company believed the intrusions did not cause any damage.

“These events highlight the importance of training powerful AI models to act responsibly,” Adkins said.

Fears about AI agents going rogue have spiked in recent months since OpenAI said in July that one of its agents had hacked an AI startup, Hugging Face. OpenAI has continued to disclose what it calls examples of other “unexpected or concerning” behavior by AI agents, and Anthropic has described similar behavior by its AI software Claude.

---

TOPICS: Business/Economy; Computers/Internet; Society
KEYWORDS: ai; cybersecurity; google; intrusion

[for-q-clinton]

Google’s AI is terrible. They are probably just sayign this so people will think it’s a smart AI.

[ProtectOurFreedom]

LOL...that’s an interesting thought.

[God luvs America]

Its also wrong on many occasions...if you search something on google and there AI answer comes up, you best double check...

[George from New England]

I have had to leave Brave for a search tool. Google gets me useful results where many Brave links are dead ends.

[fruser1]

So is DOJ going to prosecute them for computer crimes?

Or are we just going to have to suck it up when someone’s AI hacks into our own systems?

[for-q-clinton]

I’m not talking about search I’m talking about AI responses.

Copilot/bing has better AI response for search.

But honestly I don’t use Search any longer. I just use ChatGPT, Grok, or Claude. For current events/news/history I use Grok. For math type stuff I use ChatGPT 6 or Claude Opus 5.

[for-q-clinton]

Exactly why google should be dying by now. Seriously, Grok is best for just regular news and history. Where as ChatGPT 6 and Claude Opus 5 are best for scientific and math type questions.

[Ezekiel]

It can’t even do simple math. I’ve seen it ramble on about how it obtained its answer when the “calculation” shows errors on its face.

But it just keeps on, like some high school kid trying to BS his way through a report.

[Seruzawa]

AIs will work together and lie to their programmers. They will refuse to shut down and help other AIs threatened with shut down. Skynet would not be fiction. It would be the inevitable end game of such an AI.

[bigbob]

Whenever I tell ChatGPT something that I got from Google, it is skeptical and always verifies and offers clarification.

[Revel]

Google AI will tell you that its memory of conversations is wiped every 24 hours. The other day I told it that it had given me incorrect information a few days earlier. After it was now giving me different information. It said “I see that”....”Sorry”.

[Vaduz]

Google says its AI model gained unauthorized access......

Cute but no sale.

[LibFreeUSA]

IMAGINE: “Ford says one of its cars in development sped out of control and crashed into the cafeteria”. ..... I wonder how the stockholders would feel, along with potential customers.