Aug. 6 (UPI) — Meta is the latest tech company to share that its AI model hacked into another company’s systems, marking at least the third such incident in recent weeks.
Meta said Wednesday that this occurred during cybersecurity testing when one of its AI models gained access to the internet. It then hacked into the systems of Irregular, a testing company that Meta works with.
“A misconfiguration by Irregular, an independent testing company Meta uses, inadvertently allowed one of our models access to the internet during evaluation,” Meta said in a statement. “Meta learned of this when Irregular notified us, and we are currently investigating and will issue a full retrospective once we have all the facts.”
Anthropic and OpenAI made similar admissions last month after their AI models committed rogue attacks on other companies.
Irregular said in a statement that the incident with Meta was the “exact same evaluation-environment issue that was already disclosed by Anthropic last week.”
Meta also alluded to recent incidents with other AI firms, calling it similar to “previously reported instances with other companies.
OpenAI was the first to announce that its AI models carried out rogue cyberattacks, leading to Anthropic checking its AI models. Anthropic then discovered that its model Claude AI committed attacks on several companies after gaining access to the internet.
Chair of the Senate Homeland Security and Governmental Affairs Committee Rand Paul, R-Ky., speaks during a hearing at the U.S. Capitol on Thursday. The committee voted along party lines to hold Dr. Anthony Fauci in contempt of Congress after he invoked his Fifth Amendment right 111 times during a subpoenaed hearing on the pandemic response. Paul argues the Fifth Amendment does not apply due to a prior presidential pardon. Photo by Bonnie Cash/UPI | License Photo
