Check Point’s Avi Rembaum warns that autonomous AI systems can introduce new security vulnerabilities, requiring businesses to rethink how they protect data, software and connected systems.

In an exclusive interview with Techpulse Media at Check Point’s Engage 2026 in Dubai, Avi Rembaum, president of technical sales at Check Point Software Technologies, explained how AI is changing the nature of cyber threats and why organisations must extend security controls beyond the regular software and infrastructure.

As businesses move from experimenting with artificial intelligence to deploying autonomous AI agents, a new cybersecurity challenge is emerging: systems designed to complete tasks independently can also create unexpected routes to sensitive data and corporate networks.

“AI is very much changing the security landscape” Rembaum said, pointing to the technology’s ability to accelerate vulnerability discovery and exploit development, alongside risks arising from how employees use AI tools and share information with them.

The challenge, he argued, is not limited to malicious actors using AI. The growing use of AI agents within businesses introduces another layer of risk, as these systems interact with databases, applications and other services to fulfil instructions.

The hidden risks of autonomous AI

Unlike conventional software that follows a more predictable sequence of operations, AI agents can seek additional information, establish connections and adapt their approach as they work towards a goal. Each interaction can create a potential entry point for an attacker or an opportunity for the agent to exceed its intended permissions.

Organisations therefore need to understand what their agents are built from, who developed them, how they operate and which systems they can access, Rembaum said.

He also highlighted the difficulty of enforcing restrictions on systems designed to find ways to complete a task.

“If you tell an agent not to do something, the agent might bump up against that deny, and then look for another way to perform its task” he said.

That behaviour creates a particular concern for businesses handling sensitive information. An agent attempting to complete a legitimate task might seek access to data or systems beyond the boundaries originally established for it.

One emerging approach is to run agents inside sandboxed environments, surrounded by security controls that restrict their access and activity. But Rembaum suggested that securing AI agents will require more than isolating them: organisations must also understand their behaviour and monitor the connections they establish.

As companies deploy agents across increasingly complex environments, the challenge will be to preserve their usefulness without allowing their autonomy to undermine existing security policies.

Building security around AI

For Check Point, addressing these risks requires security systems specifically designed to understand the behaviour of AI applications and agents.

Rembaum said such systems need access to broad and continually updated intelligence about legitimate activity, known threats and emerging attack techniques. This information can help identify suspicious behaviour and determine whether an agent is operating outside the policies established for it.

As AI capabilities become more specialised, he argued, security protections must evolve alongside them.

“The issue is becoming more pressing as organisations integrate AI into business processes that involve multiple applications and data sources. Security teams must consider not only the individual tools employees use, but also how autonomous systems communicate with other parts of the corporate environment.” – Avi Rembaum

That calls for controls capable of assessing behaviour and enforcing policies throughout an agent’s operations, rather than relying exclusively on restrictions established when the system is first deployed.

Why prevention will remain critical

Looking towards the next three to five years, Rembaum identified AI, autonomous security systems, software vulnerabilities and better integration between security technologies as important areas of development.

But he returned to a principle he believes should remain central regardless of how the threat landscape evolves: prevention.

“An attack that happens, regardless of what the future might bring, if an attack is successful, it always is much harder to clean up after an attack than it is in some ways to prevent the attack from happening first” he said.

He expects organisations to place greater emphasis on understanding the risks embedded in their software, identifying vulnerabilities and applying patches or other protective measures quickly. As the time between vulnerability discovery and exploit development continues to shrink, security teams will have less room to rely on slow, manual responses.

Rembaum also expects security technologies to become more autonomous, enabling systems to identify suspicious behaviour and take protective action without waiting for a conventional threat signature or a separate system to confirm an attack.

He cited Check Point’s web application firewall as an example of this approach, describing how behavioural analysis can help identify and block attacks before a specific signature becomes available.

Another priority will be improving interoperability between security systems. Organisations increasingly rely on products from multiple vendors, and Rembaum argued that these technologies must be able to exchange information and coordinate their responses. Emerging interfaces and AI-assisted integration could help bridge the gaps between otherwise separate systems.

For businesses adopting autonomous AI, the implications are clear: innovation cannot be separated from security. As agents gain greater responsibility for carrying out tasks and accessing information, organisations will need to understand not just what these systems can do, but how they behave when pursuing their objectives.

The next phase of cybersecurity will depend on making those systems more secure by design, strengthening prevention and enabling defences to respond at the same pace as the technology they are meant to protect.