CrowdStrike CEO Kurtz rejects calls to slow down AI: The solution is runtime monitoring

CrowdStrike CEO George Kurtz has pushed back against the debate suggesting that the development speed of "frontier models" in artificial intelligence should be reduced. Kurtz stated that advanced and open-weight models are already widely accessible, and that slowing down development activities at this stage would not eliminate existing cybersecurity risks.

Speaking to CNBC following Anthropic CEO Dario Amodei's call for a slowdown regarding more powerful AI models, Kurtz emphasized that models with dangerous capabilities are already circulating in the public domain. Describing the situation with the phrase "the genie is out of the bottle," Kurtz believes the real need is to build security systems that can monitor what models are doing during actual use and intervene when necessary, rather than trying to stop the models themselves.

Kurtz's approach focuses on the security debate, which has become more critical with the proliferation of autonomous AI agents. While the CrowdStrike CEO acknowledges that these systems can create serious risks, he argues that the solution is not a general slowdown policy, but rather runtime controls that limit the agents' permissions, make their operations visible, and block harmful behaviors.

The discussion comes at a volatile time for tech stocks. CrowdStrike shares rising from the $206 level to over $235 on Monday also brought the company's assessments on AI and cybersecurity to the center of market attention.

TWO DIFFERENT APPROACHES IN THE SECURITY DEBATE

Kurtz's remarks coincide with growing security concerns following recent attacks associated with Hugging Face and RubyGems. These incidents have opened a debate not only on the power of AI models but also on what access and processing permissions should be granted to autonomous systems. In the same period, Microsoft's release of updates for 974 security vulnerabilities, including two zero-day flaws as part of Patch Tuesday, was among the developments highlighting the magnitude of the pressure on the defense side.

The CrowdStrike CEO stands on a similar line to David Sacks regarding the need for strict control of autonomous agents. However, he is cautious about comprehensive government interventions. According to Kurtz, overly heavy regulations could weaken the US's competitive power against its global rivals. For this reason, he believes that models where AI labs and security companies work together, moving oversight into the production environment, are more feasible.

In this context, Kurtz cited Anthropic's Project Glasswing initiative as an example. The project aims to combine lab-level protections for Anthropic's Mythos model with runtime monitoring tools in the production environment. Thus, security is not left solely to limitations during the model development phase; monitoring the model's behavior while it runs in a real environment also becomes part of the defense.

Amodei's call has also sparked debate outside the technology sector. The Trump administration opposed slowing down the pace of AI development, arguing that it could put the US at a disadvantage against China. The Chinese side also expressed the view that an approach driven by fear and fierce competition would harm global AI governance.

Thus, the debate is shaping up along two main axes: the proposal to slow down progress before more powerful models emerge, and the approach of accepting that existing models are already accessible and directing security investments toward real-time monitoring, agent controls, and defense tools. Kurtz finds the second path more realistic and applicable.

News Source: 12punto