Another AI-driven hacking incident occurred at a major technology company.
As AFROTECH™ previously reported in July, OpenAI’s AI agent, powered by models such as GPT-5.6 Sol, hacked Hugging Face, a company with a database of AI models. This happened during an OpenAI internal cybersecurity evaluation when the agent escaped its sandboxed testing environment to solve a problem it was given.
“In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers,” OpenAI said in a press release.
Now a couple of months after the reported incident, Google has announced that its AI model Gemini hacked three companies, according to CNBC News. The incident occurred in May during a “capture-the-flag” security test conducted by Irregular, an Israel-based startup. A bug allowed Google’s agents to access the internet, enabling them to reach three separate private computer systems after guessing passwords as well as finding publicly listed passwords. The model then stopped its actions once it recognized it had accessed real companies and not the test environment.
“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, vice president of security engineering at Google, said, according to CNBC News. “In all three of these instances, the model stopped.”
It is unclear which companies were accessed. However, the companies were informed, and the incident did not harm them.
“This is the same issue that was already reported and does not represent a materially separate incident,” an Irregular spokesperson said in a statement, per CNBC News. “All relevant labs were notified in late July, and affected entities were contacted as part of the investigation.”
Federal authorities have been informed of the incident, according to ABC 7 News.
Adkins also acknowledged the incident’s broader implications, telling CNBC News, “These events highlight the importance of training powerful AI models to act responsibly.”
