TORONTO — AI firm Cohere sold its services to Canada’s cyber intelligence agency, internal government documents obtained by The Logic reveal.

A memo dated October 1, 2024, states that the Communications Security Establishment (CSE) had a contract for Cohere’s technology. The briefing note had been prepared for then-prime minister Justin Trudeau ahead of a meeting with the firm’s CEO Aidan Gomez.

The company’s large language models (LLMs) can be deployed “at client premises, air gapped from the original model,” the memo states. Its technology lets developers “build products with them securely within their data environment.” The Logic obtained the memo via an access to information request.

Cohere declined to comment. CSE spokesperson Janny Bender Asselin said the agency does not comment on its contracts, citing national security.

The memo, parts of which are redacted, does not include further details about the nature of the work. Cohere’s government contract comes as Ottawa says it will significantly increase defence spending, including on AI, and as the firm is making a push into the public sector, with a focus on security. Federal lobbying records show Cohere executives communicated with Samantha McDonald, CSE’s deputy chief of innovative business strategy and research development, in July and August 2024. 

On Monday, Cohere co-founder Ivan Zhang was in the front row as Prime Minister Mark Carney announced Canada would put an additional $560 million into digital technology and cyber safeguards this year. That includes $370 million for CSE. The new money is  part of the Liberal government’s $9-billion push to meet the NATO commitment of spending two per cent of GDP on defence. 

Speaking about his presence at Monday’s announcement, Zhang told The Logic that Cohere welcomes Ottawa’s “commitment to a strong national defence and integrating the country’s technology expertise.” 

“I firmly agree that AI should be a key piece of this strategy and Cohere looks forward to doing its part to support this mission,” he added.  

In a speech on Tuesday, AI Minister Evan Solomon said defence spending is part of the Liberal government’s plan to help Canadian companies that are developing critical technologies to scale. Firms will be “validated with a government contract so they can get investment from around the world,” he said, citing similar programs in the U.S.

The Liberal government has framed Toronto-based Cohere as one of Canada’s AI champions, and the federal Strategic Innovation Fund has awarded the firm $240 million to pay for compute capacity it will buy from a new data centre that New Jersey-based CoreWeave is building in the country. 

Cohere is also trying to sell its technology to the U.S. public sector. Last week, Cohere announced a deal with Second Front Systems, a Wilmington, Del.-based firm that develops software for U.S. government departments and tech vendors. Under the agreement, Second Front will make Cohere’s AI tools available to its clients via a platform that developers use to show they comply with U.S. Defense Department requirements. Palantir, the Denver-headquartered firm which sells data analytics tools to U.S. intelligence agencies, has also reportedly made Cohere’s models available to its customers.

In February, Cohere hired David Ferris, a Washington, D.C.-based former Canadian Armed Forces infantry officer, to head its global public sector business. “AI can dramatically improve pattern recognition and anomaly detection across the vast datasets analyzed by defence and intelligence agencies,” he told the U.S. Senate cybersecurity subcommittee the following month, adding that LLMs can sort through and synthesize “huge volumes of multi-source information,” including satellite images and transcripts from intercepted communications. 

Cohere claims its technology is particularly suited for regulated industries and governments with high security standards because it can be efficiently run on customers’ own hardware, limiting the risk of losing control over sensitive data. At the committee, Ferris also said Cohere’s models operate well in multiple languages and that the firm works with clients to tune them for different fields, like intelligence. “We have done this in the commercial and national security space,” he said.

Cohere’s policies don’t explicitly prohibit military or intelligence use of its models. In a February interview with The Logic, Cohere co-founder Nick Frosst said he’s comfortable with clients from any industry employing the firm’s technology as long as they comply with its terms. Set in September 2022, the terms prohibit the use of the firm’s products to create or operate weaponry, transmit malicious code or otherwise cause death and damage. The terms also require clients to comply with data and privacy laws.

Other AI developers are also pushing into the security space. Anthropic recently launched a new set of models for U.S. national security agencies. In December, OpenAI announced a partnership with Anduril to improve the performance of the defence technology startup’s drones. Both Meta and Google have also recently updated their policies to allow the U.S. military to use their AI models.

Update: This story has been updated with details of Cohere’s lobbying of CSE officials.