SASE’s role in securing AI adoption: How existing tools can manage AI security

In a recent CRA webcast hosted by Enterprise Security Weekly podcast host Adrian Sanabria, Check Point CTO Jonathan Zanger and Check Point VP of SASE Amit Baraket discussed why SASE is uniquely positioned to reduce AI risk. The cloud-based network-security tool sits in the path between users, devices, and cloud services — exactly where AI interactions take place.

Securing AI adoption, Baraket and Zanger said, is less about whether employees will use generative AI (they will no matter what company policies state) and more about whether security teams can manage the new connectivity layer that AI creates.

Zanger framed the AI threat landscape through three lenses: how attackers use AI to supercharge traditional threats, how agentic applications change the attack surface, and how employee AI usage can introduce data loss and governance challenges.

He warned that attackers now operate "faster…at higher scale, highly personalized and with zero mistakes," turning phishing and fraud into industrialized operations.

As more apps become agentic, Zanger added, the natural-language interface becomes the target: "If that's the default interface, that's the default attack surface."

Sanabria added that threat actors once needed technical skills and target reconnaissance to pull off successful attacks, but AI has removed those barriers.

"You don't need to know what you're attacking. You can just write a prompt," he said. "You can put it in a PDF, and you can hand it to somebody, and at some point, maybe those instructions get followed."

Baraket drew a parallel to the remote-work surge during COVID, when businesses opened access to their networks to keep productivity high, and security had to catch up.

AI is similar, Baraket said, as organizations are enabling new tools, but the attack surface expands through prompt-based leakage, prompt injection, malware, and phishing.

That's where SASE helps, he explained: Because SASE already brokers user-to-app connectivity, it can discover AI usage, enforce AI governance policies and apply protection inline without relying on scattered point solutions.

Zanger and Baraket reiterated that unauthorized "shadow AI" usage by employees will be inevitable if the security team in an organization becomes the "department of no." Blocking AI sites may simply push employees to access them from personal devices.

Instead, Zanger said organizations should use SASE to whitelist approved AI tools, block risky alternatives, and coach users in real time. He positioned SASE as a control point for protecting sensitive information as prompts become conversation-like, which will encourage over-sharing as users treat AI agents as co-workers rather than machines.

"You have inline DLP for all data going to AI services," Zanger explained. "You can block PIIs, you can block source code, you can block credentials or IP."

Baraket explained how Check Point's Harmony SASE platform can find that employees are using not only major tools like ChatGPT, Gemini, and Claude, but also "hundreds of long-tail niche AI services" as well as AI features embedded inside already-approved SaaS apps.

The Harmony platform's approach, he and Zanger explained, is to classify tools, assign risk scores, and then apply identity-based access policies with different rules for different functions (e.g., allowing marketing image generation while restricting finance).

Enforcement then extends to prompt-level inspection, including redaction and user feedback designed to prevent workarounds.

AI changes quickly, Zanger and Baraket said, with agentic workflows, local models on endpoints, and new integration patterns like SaaS-to-SaaS connections all raising the stakes.

Zanger concluded that the best strategy is to adapt at the speed of AI, with a practical starting point: focus first on user behavior and zero-trust hygiene, then expand visibility and controls broadly through a unified SASE architecture.

"The short answer is that you need to use AI in order to secure AI, or secure against AI driven threats," Zanger said.