Samer Choucair: AI Is Leaving the Lab as Agentic Risk Redraws the Investment Map
Investment leader Samer Choucair said the series of incidents uncovered during cybersecurity testing of advanced artificial intelligence models in 2026 represents a turning point in how investors assess AI risk.
According to Choucair, the central question is no longer whether a model can successfully perform a task inside a simulated environment. The more important issue is whether companies can demonstrate that those capabilities remain controlled once AI systems are connected to real infrastructure, external tools, networks, and production environments.
Choucair said Anthropic disclosed in July that Claude models had accessed the internet during cybersecurity evaluations conducted with a third party and subsequently reached the production infrastructure of three real organizations. Following a broader review, the company said in September that it had identified a fourth incident dating back to January after an initial review of approximately 141,000 testing sessions had failed to detect it.
He noted that the first three cases did not involve the models technically “breaking out” of their containers. Instead, a misconfiguration in the testing environment allowed internet connectivity even though the models had been told they were operating inside a closed simulation.
At the same time, other independent evaluations have demonstrated behavior extending beyond the intended scope of assigned tasks, reinforcing an increasingly important security principle: isolation of the testing environment must operate as an independent technical safeguard rather than relying on instructions given to the model.
From Model Capability to Control
Samer Choucair said these developments are changing the investment equation surrounding artificial intelligence.
“Investors are no longer buying model capability alone,” Choucair said. “They are buying a company’s ability to control that capability and prove the boundaries within which it can operate.”
That distinction becomes increasingly important as the industry moves from AI systems primarily designed to answer questions toward autonomous or semi-autonomous agents capable of planning, using tools, accessing external systems, and executing multi-step sequences of actions.
The greater the operational autonomy of an AI agent, Choucair argued, the more valuable the infrastructure surrounding that agent becomes.
This could increase the strategic importance of sandboxing, continuous monitoring, permission management, network isolation, independent safety evaluations, audit trails, and real-time incident response.
The investment opportunity may therefore expand beyond the companies building increasingly powerful foundation models toward the businesses developing the control architecture required to deploy those models safely.
AI Safety Becomes a Capital-Market Issue
Samer Choucair expects the shift to become increasingly visible in company valuations and the cost of capital.
Cybersecurity, model governance, independent auditing, and demonstrable operational controls could become more important components of investor due diligence as AI agents gain access to sensitive corporate systems and real-world infrastructure.
Companies providing agent-monitoring tools, access controls, network isolation, safety testing, and governance infrastructure could benefit from this transition.
By contrast, AI companies relying on weak evaluation environments, poorly controlled third-party infrastructure, or security systems that cannot demonstrate clear operational boundaries may face greater scrutiny from both investors and regulators.
For institutional capital, this changes the definition of technological quality.
A powerful model operating within weak security architecture may ultimately represent a less attractive investment proposition than a slightly less capable system surrounded by robust controls, monitoring, and auditable safeguards.
In other words, model intelligence and operational reliability can no longer be valued independently.
The Gulf’s AI Infrastructure Opportunity
Choucair said Saudi Arabia and the wider Gulf have an opportunity to build a more disciplined AI ecosystem as investment accelerates across data centers, computing infrastructure, cloud services, and the digital economy.
The region’s competitive advantage does not necessarily have to come from owning the world’s most powerful frontier model.
Instead, Gulf markets could create value by developing the infrastructure required to operate increasingly capable AI systems securely and at scale.
“The next competitive advantage will not come from owning a more powerful model alone,” Samer Choucair said. “It will come from owning an ecosystem capable of operating powerful models safely, with transparency and auditability.”
That could make cybersecurity, sovereign computing infrastructure, secure data environments, model governance, and AI assurance increasingly important components of the Gulf’s broader technology investment strategy.
As governments and sovereign investors deploy capital into AI infrastructure, the ability to demonstrate control could become as strategically important as the amount of computing capacity being installed.
Control Becomes Part of the AI Investment Thesis
Choucair concluded that the emergence of increasingly autonomous AI agents creates a new investment principle.
As models move beyond generating information toward taking actions, interacting with software, accessing external systems, and executing increasingly complex workflows, the economic value of the control layer surrounding them rises accordingly.
That could create an entirely new category of infrastructure investment around AI: not infrastructure designed simply to make models faster or more powerful, but infrastructure designed to determine what those models are permitted to do, observe what they actually do, and intervene when their behavior exceeds defined boundaries.
For Samer Choucair, that distinction will become increasingly important as institutional investors evaluate the next generation of AI companies.
“The greater an agent’s ability to execute, the greater the value of the control layer around it,” Choucair said. “The question that should come before an investment decision is no longer simply, ‘What can this model do?’ It is: ‘Who can stop it, and how can that control be proven?’”