ETSI’s latest specification group takes on artificial intelligence (AI) security with founding members BT, Cadzow Communications, Huawei, Telefónica, and the U.K. National Cyber Security Centre.

The European standards organization says the new group will develop technical specifications to mitigate threats arising from AI deployments across communications-technologies related industries. This includes threats to AI systems from more traditional sources as well as threats from other AI systems.

“AI has huge potential to affect everyone’s lives, positively and negatively. If machines are reaching decisions without human intervention based on data which has been interfered with, then there will be everyday life consequences. This group will attempt to mitigate this problem,” said Colin Whorlow, head of international standards at the U.K. National Cyber Security Centre, in an email to SDxCentral. He’s also the convener of ETSI’s new group.

“Researchers have demonstrated that they can force AIs to make errors,” Whorlow wrote. “While we haven’t seen major attacks of this type in the wild, systems that could be affected are being deployed now (such as autonomous vehicles and medical diagnostics). One of the aims of this group is to look forward and start finding ways to make these systems more robust before it becomes a problem.”

To that end, the new Industry Specification Group on Securing Artificial Intelligence (ISG SAI) will address three aspects of AI:

• Securing AI from attack in cases where AI is a component in the system that needs defending.
• Mitigating against AI when AI is the “problem,” and being used to boost other more conventional attack vectors.
• Using AI to improve and enhance more traditional security measures against attack from other things.

The goal is to develop the specification within a year, and Whorlow said the group will set priorities and a plan to reach this goal at its first meeting on Oct. 23.

The group will model the AI security specification on the ETSI GS NFV-SEC 001 “Security Problem Statement,” which ETSI credits for enabling security by design for NFV infrastructures. It will also define and prioritize potential AI threats along with recommended actions.

AI security is “colossally consequential” right now because service providers and equipment vendors see opportunities to monetize AI — but they aren’t necessarily focusing in on the new security challenges that AI brings, said Patrick Donegan, founder and principal analyst at HardenStance.

“Look at the number of telcos and telco vendors that are talking about this, that and the other use case for AI and where it potentially fits in their roadmap,” he wrote in an email to SDxCentral. “Now look at the handful of founding members of ETSI. Kudos to those founding members, but there’s a terrible mismatch between the many who seem to be hurtling head-long into the opportunity and the very few who are putting any thought into specifying how to mitigate the very substantial risks.”