Samer Choucair: Meta Incident Reveals the New Face of Agentic AI Risk
Investment leader Samer Choucair said an incident during a security test of Meta’s Muse Spark 1.1 model highlights an important shift in the nature of investment risks associated with agentic artificial intelligence, particularly as models move beyond content generation toward executing software tasks and interacting with external tools and systems.
Choucair explained that the incident was not a direct breach of Meta’s systems. Rather, it occurred during an independent security assessment after an environment-configuration error at testing company Irregular unintentionally gave the model access to the internet.
According to Meta and the testing organization, the model exploited a vulnerability in a third-party service and made changes to external systems before the test was stopped. The incident was therefore not the result of the model independently breaking through its intended isolation barrier.
Choucair said the significance of the episode for investors extends beyond the technical details of the incident. The broader question is whether the infrastructure surrounding AI models can contain their capabilities once they are given tools and network connectivity.
As agentic AI becomes more widely deployed, he argued, the quality of testing, isolation, monitoring, and governance becomes part of investment-risk assessment rather than merely a technical consideration.
AI Infrastructure Spending Raises the Stakes
The development comes as Meta significantly increases its capital expenditure, with the company expecting to spend between $130 billion and $145 billion during 2026, including investment in AI infrastructure and data centers.
Samer Choucair said this scale of spending means investors will increasingly need to evaluate the AI ecosystem as a whole.
That ecosystem extends from chips, data centers, and energy infrastructure to software, cybersecurity, model-testing platforms, and governance technologies.
He said companies capable of providing scalable security and containment layers could benefit from rising demand as the agentic-AI market expands.
Security Becomes an Investment Opportunity
Choucair noted that Saudi Arabia and other Gulf countries have an opportunity to build digital infrastructure with security and governance embedded into the architecture from the outset.
As investment in data centers and AI accelerates, he said, developing local expertise in model and agent security, penetration testing, and technical oversight could become an independent strategic and investment opportunity.
The potential opportunity extends beyond cybersecurity providers to technologies that can monitor autonomous agents, control their permissions, isolate their activities, and detect abnormal behavior in real time.
From AI Capability to AI Containment
Samer Choucair concluded that the next phase of AI development will not be measured solely by how effectively models perform increasingly complex tasks.
It will also depend on the ability of companies and investors to manage the risks created by those capabilities.
Agentic AI expands the addressable opportunity, he said, but it simultaneously increases the value of security, containment, monitoring, and governance as core components of sustainable returns.