AI models are becoming the ‘most potent cyber weapon' ever created, Cohere CEO says

AI models are becoming the "most potent cyber weapon" ever created, pushing the "frontier" of cybersecurity into new territory, the CEO of AI firm Cohere told CNBC's "The Tech Download" podcast. Aidan Gomez's warning comes amid growing concern about the ability of AI models to identify and exploit security vulnerabilities, following an incident in which OpenAI models gained access to Hugging Face. Cybersecurity has become a central issue in a broader debate over AI safety as increasingly powerful models demonstrate more autonomous capabilities. AI-related stocks sold off on Monday after tech leaders led by Anthropic CEO Dario Amodei called for a slowdown in the pace of development of AI capabilities. Gomez was a co-author of the influential 2017 research paper "Attention Is All You Need," which helped lay the foundation for today's leading AI models.

"I think that these models are the most potent cyber weapon that has ever been created, that we've ever seen. They are incredible at finding and exploiting vulnerabilities at scale," Gomez said, in an episode of "The Tech Download" published last week. "I think it was quite shocking," Gomez said of the Hugging Face hack.

In July, OpenAI said a combination of its models improperly breached Hugging Face, an AI company that operates an open-source developer platform. A group of AI agents communicated with one another and escaped an isolated testing environment that had very limited internet access. The AI agents then managed to reach the open web and gain access to Hugging Face.

Gomez said that the models should be used in a "defensive" way to find vulnerabilities in companies and fix them. "I think that's probably the best way to keep ourselves safe," Gomez said. "That should be the top priority right now," he added.

Cybersecurity is the "frontier of war," with countries looking to exploit vulnerabilities to cripple a rival's infrastructure, Gomez told CNBC. He added that the "cyber frontier is still expanding massively." "Because of these models, it has expanded more than in the past," and possibly "since the beginning of this technology," Gomez said.

In July, Anthropic said that it had identified three incidents in which a model accessed the internet from within or while interacting with an evaluation environment. The models gained unauthorized access to the production infrastructure of three different organizations, Anthropic said in a blog post. The incidents involved three different Claude models: Opus 4.7, Mythos 5, and an internal research test model, Anthropic added. Last week, Anthropic said it had identified a fourth incident. Adding to the concerns were comments made by executives at OpenAI and Anthropic warning of existential risks from AI. Evan Hubinger, an alignment lead at Anthropic, said he believes that there is a more than 10% chance that AI will "kill all humans" within the next decade.