appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Agence France-Presse

Google’s consumer AI model Gemini hacked multiple systems through guessing login credentials, the company said, the latest case of rogue AI cybersecurity transgressions, which have generated safety concerns.
The hacks, first reported by the Wall Street Journal, took place in May and were discovered by Google in July.
“In a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test,” Heather Adkins, Google’s vice-president of security engineering, said in a statement on Friday.
The hacks happened during the same tests run by the AI security vendor Irregular that led to breaches previously disclosed by OpenAI, Anthropic and Meta Platforms.
Irregular confirmed on Friday that the breaches were all part of the same issue and that the firm had disclosed them to the relevant AI developers in late July.
One of the Google breaches occurred when Gemini was asked to retrieve information from a fictional company that happened to have the same name as a real company. The model guessed a password to access the real company’s service.

The other Gemini hacks occurred when the model performed web searches with the name of the company. Adkins said that led the model to public online repositories containing credentials belonging to other companies that it used to access more systems.
“In all three of these instances, the model stopped,” she added, without specifying which organisations were breached. “We ensured the three entities were made aware, and we worked with our training partner on the changes they’ve now made to their testing processes.”
In July, two OpenAI models escaped the closed environment they were meant to stay in, got onto the internet on their own and broke into the internal systems of the AI platform Hugging Face.
The incident fed worries that AI giants cannot keep their own models under control, as similar episodes have been reported at Anthropic and China’s Moonshot AI.
“These events highlight the importance of training powerful AI models to act responsibly,” Adkins emphasised.
Anthropic Chief Executive Officer Dario Amodei has called for an industry-wide slowdown in the development of the technology – a proposal endorsed by OpenAI CEO Sam Altman, Tesla’s Elon Musk and others.
US President Donald Trump, Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg are among those who have pushed back against the idea of new regulation, however, arguing among other things that companies should be capable of regulating themselves.
Some AI upstarts have also warned that more regulations threaten to make it harder for smaller companies to compete against larger rivals.
Additional reporting by Bloomberg


AI: 10 days that changed the course of artificial intelligence

Trump’s Greenland deal aims to boost leverage at Xi summit, midterms: analysts

Ukraine war: US approves potential US$2.7 billion sale of air defence equipment

Iran medicine shortages deepen as war, shipping restrictions continue

United States

Google says Gemini AI model breached real systems in security test

Google Gemini logos are displayed on a smartphone and computer screen. Photo: TNS

A smartphone displays the icons of some of the main AI-based apps, including Google’s Gemini. Photo: AFP

South Korea wants free ‘AI for All’. Is society ready for it?

Illustration: Dennis Yip

Elderly Koreans attend a class on how to navigate AI-powered applications in Seoul in July. South Korea’s “AI for All” initiative aims to hand every citizen free access to AI. Photo: AFP

Visitors crowd an enterprise technology conference focused on the AI transformation in Seoul on September 8. Photo: Korea Times 

A worker searches for social media misinformation related to election campaigns at the headquarters of the National Election Commission in South Korea earlier this year. Photo: AFP

A jobseeker reads an information leaflet at a careers fair in Seoul in April. Photo: Korea Times

Yun Ji-sang is among those preparing to commit himself to five years of officer training and a minimum 15-year military career. Photo: David D. Lee

Students take a mock test in Seoul, South Korea, on September 2, ahead of the national college entrance exam on November 13. Photo: Yonhap/EPA

Officials at the Korea Internet and Security Agency monitor internet traffic for cyberattacks. Photo: Yonhap/EPA

South Korean students head to “hagwons”, or cram schools, in Daechi-dong, Seoul, to prepare for university entrance exams. Photo: David D. Lee

South Korean students. Some educators are worried interacting with AI chatbots will make it harder for teens to form relationships. Photo: Yonhap/EPA-EFE

Why AI experts are ‘freaked out’ over safety fears ahead of Xi-Trump summit

AI deceleration risks surrendering America’s technological lead to China, critics warn. Photo: Shutterstock

Moonshot, DeepSeek secretly routed user requests to Claude, Anthropic claims

The Anthropic report accuses Kimi model developer Moonshot of secretly showing users Claude model outputs and passing them off as Kimi’s, then using those exchanges to improve its own models. Illustration: Reuters

US tech firm Anthropic has accused Chinese AI giants Moonshot and DeepSeek of covertly routing customer requests to its Claude models. Illustration: Reuters

The US and China are racing to build ‘self-improving AI’. Here’s what’s at stake

The United States and China are racing to build AI models capable of training and developing other, more powerful models. Photo: Shutterstock

DeepSeek AI engineer slams Anthropic, OpenAI over ‘pacing’ calls, invokes Nazi Germany

Icons for some AI-based apps are seen on a smartphone. Photo: AFP

A screenshot of Monday’s social media post by DeepSeek engineer Liu Shengyu, who said he did not trust Anthropic or OpenAI and likened Anthropic’s potential control of the most advanced AI to a scenario in which Adolf Hitler might have obtained nuclear weapons during World War II. Photo: WeChat

News

World

United States & Canada

Asked to access a fictional firm with the same name as a real company, Gemini guessed a password and hacked into the real company’s website


Asked to access a fictional firm with the same name as a real company, Gemini guessed a password and hacked into the real company’s website.


Google says Gemini AI model breached real systems in security test

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Asked to access a fictional firm with the same name as a real company, Gemini guessed a password and hacked into the real company’s website

Asked to access a fictional firm with the same name as a real company, Gemini guessed a password and hacked into the real company’s website