Bell, Cohere release AI model designed to aid cybersecurity investigations

Bell Canada and Cohere Inc. are deploying a new artificial-intelligence model designed to assist with cybersecurity investigations as organizations around the world look to harness the technology’s defensive capabilities.

The customized large language model was built with Cohere’s technology and trained using data and expertise from Bell Cyber. The telco unit provides security services to customers ranging from small businesses to large companies such as Canadian Imperial Bank of Commerce and the Canadian and Ontario governments.

Organizations across the globe have been racing to integrate AI into their cybersecurity defences in response to growing concerns about how hackers can use the technology.

“When you’re being attacked by AI, you need AI to defend yourself,” John Menezes, president and chief executive officer of Bell Cyber, said in an interview. “AI works so fast, is doing things so fast that if I don’t have AI working for me, I would be dead meat.”

Bell spent seven months training the Cohere model, feeding it nearly a decade’s worth of data so it can analyze and summarize complex security information and provide the context needed for investigations.

That frees up analysts to focus on higher-level tasks such as validating threats, determining their impact and responding to them, the company said.

Canada’s plan to counter the U.S. and China in AI is taking shape. Will it work?

Bell has integrated the model into its autonomous Security Operations Centre, or SOC. The term refers to the nerve centre of an enterprise’s cyber defences. Autonomous SOCs leverage AI and automation to detect, triage and contain cyber threats.

Concerns about the cybersecurity risks posed by AI have been amplified by recent incidents involving agents breaking into companies’ systems during testing. These have led to calls for a co-ordinated slowdown in the technology’s development.

In May, Google’s AI model Gemini hacked into three companies during an evaluation of its cybersecurity capabilities. That incident, which the company confirmed earlier this month after a report from The Wall Street Journal, predated the July intrusion of AI firm Hugging Face by a swarm of OpenAI’s agents, which broke out of their test environment to cheat their way through an exercise.

Anthropic has experienced similar incidents. The San Francisco-based AI firm recently disclosed that its Claude models had hacked into third-party systems during testing on four occasions.

The model announced by Bell and Cohere on Wednesday is the latest development in a continuing collaboration between the two companies. Last year, they announced plans to sell AI tools to governments and businesses.

Under that deal, Bell said it would sell access to Cohere’s software platform, North, to its own enterprise and government customers. The platform enables companies to build custom AI agents to automate their workflows. Bell also said it would host the technology on its data centre infrastructure, offering Canadian companies an alternative to U.S. cloud computing giants.

Vass Bednar: The dangers of letting AI agents do your shopping

The new cybersecurity model is running on Bell AI Fabric data centres in British Columbia, which keeps sensitive security information within Canada, the company said. (Bell AI Fabric is the company’s artificial-intelligence infrastructure division.)

“Basically, what we have created is a sovereign autonomous security operations centre,” Mr. Menezes said.

The risk of relying on American technology was highlighted earlier this year, when the U.S. government directed Anthropic to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals, citing national-security concerns.

“We had access to one of the frontier models on the Friday morning, and we were so excited because we got access to the frontier model, and Friday evening, access was gone,” Mr. Menezes said.

“Canada can’t function when somebody can just, you know, shut off the power,” he added.