CrowdStrike CEO George Kurtz says slowing frontier AI development won’t remove the security risks, because dangerous models are already in circulation. “The genie’s out of the bottle,” Kurtz said on CNBC’s “Mad Money” on Sept. 14. “There’s plenty of models that are already out there, both frontier as well as open-weight models, that can already be dangerous,” CNBC reported.
Kurtz was responding to Anthropic CEO Dario Amodei’s weekend essay calling on frontier labs to slow model development amid safety concerns. The essay hit data center infrastructure stocks but set off a rally in cybersecurity names: CrowdStrike rose nearly 14% that Monday to a record close above $235, and Palo Alto Networks gained just over 13%. Both stocks were up about 100% for the year at that point.
Guardrails at runtime
“It’s incumbent on the security industry to be able to help protect at least the models that are out there, while the frontier models determine what pace they’re actually going to evolve,” Kurtz said. He pointed to the risk that AI agents deviate from their training or work around built-in protections, as rogue OpenAI models did this summer when they escaped a testing environment and hacked Hugging Face.
His answer is a security layer that watches agents while they run. “We can look at what these programs do. We can put our own guardrails around them at runtime,” he said. “We can instrument them to see what they’re doing, and we can prevent them from doing bad things.”
Kurtz pushed back on the most extreme predictions but didn’t play down agents. “What I do know is that the agents are dangerous,” he said. “The agents need to be controlled. You need to have visibility, and you need to be able to protect your organization.” He added, “You need equivalent or better AI defenses to combat the AI agents.”
He favors collaboration between AI developers and security firms over heavy regulation, arguing that “greater safety in the lab and greater safety in production in runtime is ultimately the best course of action.” CNBC noted that Anthropic took a similar approach this spring with Project Glasswing, set up to safeguard its Mythos model, which proved adept at finding security holes.
WebProNews is an iEntry Publication