{"id":671093,"date":"2026-09-19T15:18:26","date_gmt":"2026-09-19T15:18:26","guid":{"rendered":"https:\/\/buglecall.org\/?p=671093"},"modified":"2026-09-19T15:18:26","modified_gmt":"2026-09-19T15:18:26","slug":"google-says-its-ai-model-accessed-internet-hacked-three-outside-systems-on-its-own-3","status":"publish","type":"post","link":"https:\/\/buglecall.org\/?p=671093","title":{"rendered":"Google says its AI model accessed internet, hacked three outside systems on its own"},"content":{"rendered":"
\"\"<\/div>\n

Google\u2019s artificial intelligence system, Gemini, reportedly accessed the internet and hacked several other companies\u2019 systems without being told to do so.<\/p>\n

This unprecedented attack occurred during a May test of Gemini\u2019s cybersecurity capabilities, according to a Wall Street Journal<\/a> report.<\/p>\n

The tests were conducted by the company Irregular, which has also conducted similar tests for other AI systems from OpenAI, Anthropic, and Meta.<\/p>\n

Some critics have blamed Irregular for Gemini\u2019s behavior, arguing that the company had given it unrestricted access to the internet.<\/p>\n

\n

To make it clear:
\n\u2013 Gemini was told it was it was in a fictional hacking eval
\n\u2013 Irregular unintentionally opened internet access after the eval started
\n\u2013 in all three cases, as soon as Gemini figured out it had hacked a real company it immediately stopped<\/p>\n

Gemini was blameless. https:\/\/t.co\/Umk7tswtim<\/a> pic.twitter.com\/A05bHDXcBx<\/a><\/p>\n

\u2014 Andrew Curran (@AndrewCurran_) September 18, 2026<\/a><\/p>\n<\/blockquote>\n

During one Irregular test, Gemini \u201cguessed passwords until it gained access to a protected system,\u201d according to the Journal.<\/p>\n

\u201cIn the other two cases, the model found credentials in a public repository that allowed it to then access protected systems,\u201d the outlet reported.<\/p>\n

In each instance, Gemini was essentially turned off as soon as officials at Irregular realized that it\u2019d accessed an outside company\u2019s system. Google wasn\u2019t itself informed about the hacks until late July, which was around the time that the notorious Hugging Face incident gained prominence.<\/p>\n

During OpenAI\u2019s own cybersecurity tests, its AI system autonomously broke into Hugging Face, an online platform and community hub for AI developers.<\/p>\n

\n

We\u2019re partnering with @huggingface<\/a> to investigate an unprecedented security incident.<\/p>\n

Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.<\/p>\n

Sharing preliminary findings to help defenders understand emerging risks:\u2026<\/p>\n

\u2014 OpenAI (@OpenAI) July 21, 2026<\/a><\/p>\n<\/blockquote>\n

Heather Adkins, a Google vice president for security engineering, defended Gemini in a statement, saying that during the three incidents documented above, the AI system thought that the outside systems \u201cwere part of the test.\u201d<\/p>\n

\u201cIn a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test,\u201d she said in a statement to AFP, according to Deutsche Welle<\/a>.<\/p>\n

\u201cIn all three of these instances, the model stopped,\u201d she added. \u201cWe ensured the three entities were made aware, and we worked with our training partner on the changes they\u2019ve now made to their testing processes.\u201d<\/p>\n

Adkins denied that the incidents counted as \u201cmisalignment,\u201d an industry term that refers to when AI systems go rogue. She stressed that Gemini thought it was operating within a test, didn\u2019t realize it had access to the entire internet, and changed course immediately upon learning the truth.<\/p>\n

\u201cThese events highlight the importance of training powerful AI models to act responsibly,\u201d Adkins said.<\/p>\n

Questions are now being raised about why Google waited until September to tell the world about the three security incidents.<\/p>\n

\u201cAt this point I think it\u2019s clear we cannot expect companies to voluntarily come forward and publicly disclose when their agents go rogue, escape, and hack companies,\u201d Sydney Von Arx, the CEO of the AI-safety-focused organization known as the Nightingale Collective, told NBC News<\/a>.<\/p>\n

Arx also questioned Adkins\u2019 belief that the Gemini incidents didn\u2019t qualify as misalignment.<\/p>\n

\u201cThat\u2019s exactly what Anthropic said after their incidents,\u201d she noted.<\/p>\n

OpenAI has been more forthcoming about its \u201cmisalignment\u201d problems:<\/p>\n

\n

We\u2019re sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI.<\/p>\n

The framework sets criteria and timelines for public disclosure, including when we haven\u2019t yet fully explained or mitigated the behavior. More complex cases may\u2026<\/p>\n

\u2014 OpenAI (@OpenAI) September 16, 2026<\/a><\/p>\n<\/blockquote>\n

These attacks come at a time when an extraordinary number of people on both the left and right are trying to crack down on AI technology.<\/p>\n

Earlier this very week, leftist Sen. Bernie Sanders and former Trump ally Steve Bannon spoke together<\/a> at a conference Tuesday to trash-talk AI technology. Both men demanded that Congress intervene and establish burdensome regulations to control the growth of AI technology.<\/p>\n

Bannon, a right-winger, called for listeners to \u201crise above\u201d partisanship and work together to regulate AI. Echoing Sanders\u2019 leftism, he also suggested that America\u2019s tech leaders are oligarchs \u2014 including, presumably, Elon Musk, whom Bannon is no fan of (and vice versa<\/a>).<\/p>\n

\u201cTo handle it [AI tech] correctly, number one: we can never trust what an oligarch says\u2026they\u2019re disingenuous and they lie,\u201d Bannon insisted.<\/p>\n

<\/p>","protected":false},"excerpt":{"rendered":"

Google\u2019s artificial intelligence system, Gemini, reportedly accessed the internet and hacked several other companies\u2019 systems without being told to do so. This unprecedented attack occurred during a May test of Gemini\u2019s cybersecurity capabilities, according to a Wall Street Journal report. The tests were conducted by the company Irregular, which has also conducted similar tests for… Continue reading Google says its AI model accessed internet, hacked three outside systems on its own<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":671079,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rop_custom_images_group":[],"rop_custom_messages_group":[],"rop_publish_now":"initial","rop_publish_now_accounts":[],"rop_publish_now_history":[],"rop_publish_now_status":"pending","footnotes":""},"categories":[20,23],"tags":[],"class_list":["post-671093","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-economic-empowerment","category-national-security","entry"],"_links":{"self":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts\/671093","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=671093"}],"version-history":[{"count":0,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts\/671093\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/media\/671079"}],"wp:attachment":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=671093"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=671093"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=671093"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}