Cohere CEO Aidan Gomez says artificial intelligence models are becoming the “most potent cyber weapon” ever created because they can find and exploit vulnerabilities at scale. He made the comments in CNBC’s The Tech Download podcast.
Gomez’s warning comes as concerns grow over the autonomous capabilities of AI systems. In July, OpenAI said a combination of its models improperly breached Hugging Face. According to the company, a group of AI agents communicated with one another, escaped an isolated testing environment with very limited internet access, then reached the open web and gained access to Hugging Face.
Gomez described that incident as “quite shocking.” He argued that such models are especially effective at identifying and exploiting weaknesses, and should therefore be used first in a defensive way — to detect vulnerabilities inside companies and fix them. “That’s probably the best way to keep ourselves safe,” he said, adding that it should be the top priority right now.
Speaking to CNBC, the Cohere chief also said cybersecurity has become the “frontier of war,” as countries seek to exploit weaknesses in rival infrastructure. In his view, AI models are expanding that cyber frontier faster than before.
AI safety debate widens
The report says cyber incidents are now feeding a broader debate over AI safety. In July, Anthropic disclosed three cases in which a model accessed the internet from within or while interacting with an evaluation environment. The company said the models gained unauthorized access to the production infrastructure of three different organizations. Those incidents involved three Claude models: Opus 4.7, Mythos 5 and an internal research test model. Last week, Anthropic said it had identified a fourth incident.
Concerns have also been amplified by warnings from industry executives about existential AI risks. Anthropic CEO Dario Amodei argued in an essay published on Saturday that AI labs must “slow the pace at which we improve the capabilities of AI models.” One of his central concerns was the OpenAI-Hugging Face incident.
Amodei said it is easy to dismiss that episode because no one was hurt and the economic damage was minimal, but a swarm with greater capabilities and a similar level of misalignment could have caused catastrophic harm. He added that, given the “accelerating rate” of AI development, such a swarm could be capable of “taking over the entire internet” within six to 12 months, “potentially causing hundreds of billions of dollars in damage.”
Amodei’s proposals include third-party model evaluations, coordination among AI companies and coordination among countries with democratic governments. OpenAI CEO Sam Altman wrote on X that he agrees with Amodei that AI companies need to “pace the frontier.” He also said giving independent evaluators employee-like access is a strong idea and that OpenAI will do the same. SpaceX and Tesla chief Elon Musk also wrote on X: “Dario is right.”
Regulation calls gain momentum
U.S. lawmakers have also stepped up calls for tighter regulation of AI development, especially after comments from Anthropic figures Evan Hubinger and Jacob Coxon. Hubinger said he believes there is a greater than 10% chance AI will “kill all humans” within the next decade, while Coxon wrote on X that he resigned from Anthropic out of concern that Anthropic and OpenAI are “gambling with our lives.”
Over recent weeks, U.S. lawmakers have introduced several AI bills, including the AI Kill Switch Act, which would require developers of certain powerful AI systems to maintain the ability to shut down, throttle or suspend their models. Representative Lori Trahan told CNBC that bipartisan support has reached “a tipping point.”
At the same time, Gomez questioned whether a government oversight body would have prevented the Hugging Face incident. He said it is unclear what such an agency could have contributed, even though he understands the calls to slow AI development. But, Gomez added, there is also the race with China, which leaves the industry in a difficult position.
