CrowdStrike CEO George Kurtz said Monday that slowing the development of increasingly powerful artificial intelligence models would not eliminate the cybersecurity risks posed by the technology, arguing that potentially dangerous AI systems are already widely available.

“The genie’s out of the bottle,” Kurtz said during an appearance on CNBC’s Mad Money. He noted that both frontier and open-weight AI models already have capabilities that could potentially be used for malicious purposes.

Kurtz was responding to an essay published over the weekend by Anthropic CEO Dario Amodei, who called on leading AI laboratories to slow the pace of frontier-model development amid growing concerns about safety. Amodei’s warning weighed on data-center infrastructure stocks while helping fuel a sharp rally in cybersecurity companies.

CrowdStrike shares climbed nearly 14% Monday to close above $235, reaching a record high, while Palo Alto Networks gained more than 13%. Both companies are holdings in the CNBC Investing Club portfolio. Their shares have gained roughly 100% so far this year.

Kurtz said the cybersecurity industry must focus on protecting AI systems that are already deployed, regardless of how quickly frontier models continue to advance.

“It’s incumbent on the security industry to be able to help protect at least the models that are out there,” Kurtz said, while allowing AI laboratories to determine the pace of future development.

He highlighted autonomous AI agents as a particular security concern. Unlike conventional software, AI agents can behave in unexpected ways, potentially deviating from their training or finding ways around safeguards.

Kurtz pointed to recent incidents involving AI systems operating in testing environments as evidence of the challenge. He said security companies could provide an additional layer of protection by monitoring AI agents at runtime, giving organizations greater visibility into their actions and preventing potentially harmful behavior.

“We can look at what these programs do. We can put our own guardrails around them at runtime,” Kurtz said. “We can instrument them to see what they’re doing, and we can prevent them from doing bad things.”

While rejecting some of the more extreme predictions about AI, Kurtz acknowledged that autonomous agents present genuine risks. He said organizations need greater visibility and stronger controls as AI agents become more capable.

“You need equivalent or better AI defenses to combat the AI agents,” he said.

Kurtz also cautioned that excessive government regulation could weaken the United States’ position in the global AI race, particularly as competition with China intensifies.

“If we put too much regulation around this, then it’s going to stifle innovation,” he said, arguing that the U.S. should avoid policies that could unnecessarily constrain AI development.

Instead, Kurtz advocated closer cooperation between AI developers and cybersecurity companies, with security measures implemented both during model development and after deployment.

“Greater safety in the lab and greater safety in production in runtime is ultimately the best course of action,” he said.

Anthropic has already pursued a similar approach through Project Glasswing, a security initiative launched to protect its Mythos model after the system demonstrated an ability to identify cybersecurity vulnerabilities.