“If a human was doing these tasks, we’d at least have some time to detect it, respond, but the agents are so fast, by the time we know about it the damage is already done.”

Image:

Adam Ely, general manager, AI Security, Check Point Software

Enterprise adoption of AI is outpacing governance controls and CISOs are struggling to catch up and plug attacks. What’s worse, AI may be operating the attacks instead of just executing malicious commands.

“Companies are more hyper connected than ever before. Agents are acting, executing, and completing tasks faster than humans can. Therefore, we can’t even respond fast enough,” said Adam Ely, general manager, AI Security at Check Point Software.

“[We] hear from the security operations center and incident response teams [that] if a human was doing these tasks, we’d at least have some time to detect it, respond, and try to prevent it, but the agents are so fast, by the time [we] know about it the damage is already done.”

Through customer feedback and research from the past year, Check Point has documented intrusions in which AI ran exploitation workflows autonomously, generating thousands of executed commands across dozens of sessions with minimal human direction between steps.

In one breach of nine Mexican government agencies, a single operator ran two commercial AI tools together, Claude Code to break in and explore networks and GPT-4.1, generating 5,317 AI-executed commands across 34 attack sessions, to analyze stolen data and task follow-on activity.

Additionally, AI can now turn a fresh vulnerability disclosure into a working exploit within hours, according to its annual AI security report.

“What I hear from CISOs is there’s this bifurcation [of AI] happening in most enterprises,” said Ely during a media roundtable at Check Point Engage Singapore 2026.

“These AI agents are running in lots of different places. They’re running on people’s laptops, in their own data center, in the cloud service providers. They’re hosted on SaaS platforms.

“Companies are saying, how do I control this? […] How do I get my hands around everywhere? And the problem is going to get worse.

“Companies have an incomplete picture of what is actually happening. They might be protecting an agent from an attack, but they are not watching and able to protect what the agent is doing [beyond] one or two steps.”

What Check Point is doing to keep up to speed

Like many security vendors trying to be extra steps ahead and support customers, Check Point has turned internally for solutions.

“We collect lots of information about AI attacks and good AI usage,” shared Ely. “We have our own foundational model that we developed that we train and maintain. This is built on live intelligence.

“It’s built on traffic to the applications that we protect. Some of our customers allow us to use some of the signals — it’s a configuration customers can turn on or off.

“This foundational model has been what builds the small models for all of our products.”

However, as with every AI-related issue, Ely believes the core principles around managing security risks remain the same. Topmost being the importance for cybersecurity leaders and teams to partner and integrate with the rest of the organization. This enables an understanding of the intent of the company’s adoptions and applications. It is also critical to have regular conversations of what risk looks like at all levels.

“Deep technical integration is very important in this space,” said Ely. “Ensuring that companies are putting their protections for AI really in line with the AI applications or at the runtime level of the application is really important. As models change or functions change, that implementation can then be built on versus having to do a completely new security implementation.”