Benzinga and Yahoo Finance LLC may earn commissions or revenue on some articles through the links below.
Artificial intelligence is advancing so quickly that it could soon go beyond human understanding, former Google CEO said Eric Schmidt.
“An extraterrestrial intelligence is coming,” he said at the Sifted Summit in London earlier this month, raising concerns that AI models could be hacked, repurposed and weaponized in ways beyond human control.
Schmidt, who led Google from 2001 to 2011, described how the rapid progress of AI also exposes deep security flaws. “There is evidence that you can take models, closed or open, and hack them to remove the railings,” he said.
Don’t miss:
He said such manipulation could allow the systems to generate dangerous information: “a bad example would be if they learned how to kill someone.” Schmidt said major developers build in protections, but he said “there is evidence that they can be reverse engineered,” he said at the Sifted Summit.
He said hackers exploit vulnerabilities through flash injections and jailbreaking. According to Microsoft (NASDAQ:MSFT), fast injection attacks manipulate AI systems by feeding them malicious input disguised as legitimate user messages, causing the model to produce unwanted responses.
Jailbreaking, for its part, is a technique that, according to Microsoft, “can cause security barriers to fail (mitigations),” allowing AI to ignore security restrictions and generate prohibited or harmful content. Schmidt said these risks show why the world lacks an effective “nonproliferation regime” to stop the misuse of AI, a framework similar to nuclear arms control that does not yet exist.
Trending: The Microsoft Climate Innovation Fund Just Backed This Farmland Manager: And accredited investors can join the same fund
“The arrival of an extraterrestrial intelligence that is not us and that is more or less under our control is a big problem for humanity,” Schmidt said.. He added that, despite his warnings, AI remains greatly underestimated, a view that he and former US Secretary of State Henry Kissinger achieved in their joint research.
Schmidt said the thesis is proving true as the machines’ capabilities “far exceed what humans can do over time.” He cited OpenAI’s ChatGPT, which reached 100 million users in just two months, as an example of the rapid adoption of AI. “I think it hasn’t been overstated, it hasn’t been overstated,” he told the Sifted Summit audience.