AI hacking will be more common if changes aren’t made, scientists warn

Dozens of Canadian academics are among an international group of scientists calling for better containment measures for AI models.

“Multiple organizations have disclosed high-profile security failures: their AI models bypassed insufficiently sandboxed environments to compromise the systems of external organizations,” they say in an open letter.

The letter, released Tuesday, says that unless immediate changes are made to how AI is developed and deployed, “such incidents will continue to occur, and future incidents will be far more damaging to our exposed critical infrastructure.”

They say there is a need for more research on AI safety as well as emphasis on containment technologies and procedures. The scientists also want to see an international body that standardizes best practices.

“In the short term, strong sandboxing mechanisms including airgapping (disconnecting from the internet) should be used to provide high-assurance containment of cyber-capable AI models,” the letter says.

The Canadian signatories to the letter include AI pioneers Yoshua Bengio and Geoffrey Hinton, as well as Nicolas Papernot and Catherine Régis, who co-lead the research program at the Canadian AI Safety Institute. Johns Hopkins University professor Gillian Hadfield, who was recently named to Prime Minister Mark Carney’s new AI advisory council, is also on the list.

Reports of rogue AI agents have helped fuel widespread public alarm about AI safety. OpenAI disclosed over the summer that its AI agents hacked AI startup Hugging Face without being instructed by any human to do so.

That was followed by other disclosures. Anthropic said its artificial intelligence models hacked into three other organizations during testing. Meta disclosed that one of its AI models accessed the internet on its own and hacked another company. And Google confirmed its Gemini AI model hacked three companies as part of a test of its cybersecurity capabilities.

A San Francisco-based lab that monitors AI activity said in a report on Sept. 30 that the website for Library and Archives Canada was one of the targets of a failed hacking attempt by an artificial intelligence agent.

Artificial Intelligence Minister Evan Solomon said in September he spoke with his international counterparts in the G7 about setting up a “technology stability board” that would evaluate AI models for testing.