Scientists call for stronger containment of AI models after hacking incidents
An open letter signed by Canadian AI pioneers warns of future damage to critical infrastructure without immediate changes.
OTTAWA — Dozens of Canadian academics have joined an international group of scientists calling for stronger containment measures for artificial intelligence models after a series of high-profile security failures.
"Multiple organizations have disclosed high-profile security failures: their AI models bypassed insufficiently sandboxed environments to compromise the systems of external organizations," they said in an open letter released Tuesday.
The letter warns 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."
READ MORE: OpenAI discloses six cases of AI misbehavior as industry grapples with safety concerns
READ MORE: Bell and Cisco partner to build sovereign AI infrastructure for Canada
Canadian signatories 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.
The scientists said there is a need for more research on AI safety as well as emphasis on containment technologies and procedures.
They 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.
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.
With files from The Canadian Press, BNN Bloomberg and Narcity