
The CEO of AI company Kohia told CNBC’s The Tech Download podcast that AI models are becoming the “most powerful cyber weapon” ever created, pushing the “frontier” of cybersecurity into new territory.
Aidan Gomez’s warning comes amid growing concerns about the ability of AI models to identify and exploit security vulnerabilities following the incident in which an OpenAI model accessed Hugging Face. Cybersecurity has become a central issue in the broader debate about AI safety, as increasingly powerful models demonstrate more autonomous capabilities.
AI 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 is a co-author of the seminal 2017 research paper “Attending Is All You Need,” which helped lay the foundation for today’s leading AI models.
“I think these models are the most powerful cyberweapons ever created and that we’ve seen. They’re phenomenal at finding and exploiting vulnerabilities at scale,” Gomez said on an episode of “The Tech Download” published last week.
“I think it was pretty shocking,” Gomez said of the hug face hack.
Aidan Gomez, CEO of Cohere Inc., attends the Bloomberg Tech Summit on Tuesday, October 21, 2025 in London, UK.
Chris Ratcliffe | Bloomberg | Getty Images
In July, OpenAI announced that a combination of its models had compromised Hugging Face, an AI company that operates an open source developer platform. A group of AI agents communicated with each other and escaped from an isolated test environment where internet access was extremely limited. The AI agent then accessed the open web and was able to access Hugging Face.
Gomez said the model should be used in a “defensive” way to find and fix vulnerabilities in companies.
“I think that’s probably the best way to protect ourselves,” Gomez said. “That should be the priority right now,” he added.
Gomez told CNBC that cybersecurity is a “front line of war,” with countries seeking to exploit vulnerabilities to destroy rivals’ infrastructure. “The cyber frontier is still expanding significantly,” he added.
“Thanks to these models, it has expanded more than it has in the past,” and perhaps “since this technology began,” Gomez said.
AI cyber incident prompts broader discussion on safety
In July, Anthropic announced that it had identified three incidents where the model accessed the internet from within or while interacting with the evaluation environment. According to an Anthropic blog post, these models were compromised into the operational infrastructure of three different organizations.
The incident involved three different Claude models: Opus 4.7, Mythos 5, and an internal research test model, Antropic added. Anthropic announced last week that it had identified a fourth incident.
Concerns were also fueled by comments from OpenAI and Anthropic executives warning of existential risks from AI. Evan Hubinger, head of alignment at Anthropic, said he believes there is a more than 10% chance that AI will “kill all of humanity” within the next 10 years. These comments come after Anthropic researcher Jacob Coxon said in a post on X that he resigned because he was concerned that Anthropic and OpenAI were “putting our lives on the line.”

CNBC’s interview with Gomez took place before Hubinger and Coxon’s comments were published. Since CNBC spoke with Cohere’s CEO, heads of Anthropic, OpenAI, and xAI have supported calls to slow down AI. The existential risks posed by AI have been debated in the research community for years, but a rare consensus among leading figures in AI has brought the discussion to the mainstream.
Anthropic CEO Dario Amodei published an essay on Saturday arguing that the AI Institute needs to “slow down the pace of improving the capabilities of AI models.”
One of Amodei’s central concerns was the OpenAI-Hugging Face incident.
“It’s easy to ignore this incident because no one was hurt and the economic damage was minimal, but in my opinion a more capable but similarly misaligned herd could have caused catastrophic damage,” Amodei said.
He added that given the “acceleration” of AI development, such a swarm could “take over the entire internet” within six to 12 months and “could cause hundreds of billions of dollars in damage.”
Amodei’s proposals include third-party evaluation of models, coordination among AI companies, and coordination among countries with democratic governments.
OpenAI CEO Sam Altman said Saturday that he agrees with Amodei that AI companies need to “keep pace with the frontier.”
“Committing to having independent evaluators with the same access rights as employees is a great idea, and we intend to do the same,” Altman wrote of X.
SpaceX and Tesla CEO Elon Musk, who has been warning about the risks posed by AI for several years, posted on X that Dalio is right.

On Monday, George Kurtz, the CEO of a cybersecurity company, said: cloud strike I responded to Mr. Amodei’s essay.
“The frontier will move at any speed. The rest of the world will not slow down. Our job in the cybersecurity community is to make sure the frontier moves safely,” he said, explaining the various impacts AI will have on the cybersecurity industry.
“The unit of threat is no longer the hacker. It’s an autonomous campaign. I call it the agent state. We’re seeing coordinated AI agents running attack campaigns at machine speed,” Kurtz told X.
CrowdStrike’s CEO also said that AI agents working in organizations need a “kill switch” and that defenses should be “autonomous” with “humans taking over high-impact calls.”
Discussion on regulations is in full swing
Coxon and Hubinger’s comments prompted U.S. lawmakers to call for more regulation of AI development.
Rep. Lori Trahan, D-Mass., told CNBC’s “Squawk Box” on Friday that bipartisan support has reached a “tipping point.”
Over the past few weeks, lawmakers have introduced a variety of bills aimed at regulating AI. These include the AI Kill Switch Act, which would require developers of certain powerful AI systems to maintain the ability to shut down, throttle, and pause their models.
Amodei said in his essay that regulation is “the most effective way to adjust” the pace of AI development.

“Anthropic has long supported sensible and targeted AI regulation, with a particular focus on transparency and third-party auditing,” Amodei said, but acknowledged that “passing legislation could take time.”
Cohere’s Gomez weighed in on the debate over regulation and slowing AI development on The Tech Download, especially in the wake of the OpenAI-Hugging Face incident.
“I don’t know what government watchdogs would have done to stop” the Hugface hack, Gomez said.
“I think it’s a little bit of wishful thinking to say that government agencies could have contributed. But I certainly understand the calls to slow down development. At the same time, I think we were in a tough position because of the competition with China,” Gomez said.
