UAE AI cyber risks rise as adoption accelerates

The report also found that 63% of UAE organisations do not have a complete inventory of their AI agents, while 53% are not well prepared to contain and recover from unintended agent actions. Another 91% said recovery plans need moderate or significant change for AI-accelerated attacks, underscoring that governance alone is not enough if a system can still make a harmful change at machine speed.

That matters economically because downtime, data corruption and credential compromise hit revenue, working capital and operating margins long after the initial incident. In sectors tied to finance, energy, healthcare, logistics and government services, the cost of a failed AI deployment is not just a security issue — it can interrupt transactions, delay services and raise compliance costs across the enterprise.

For investors, the message cuts both ways. Companies selling AI infrastructure, cybersecurity and data-resilience tools may see stronger demand as chief information officers move from experimentation to hardening their environments, while firms with weak controls face higher breach risk, heavier remediation bills and slower AI rollouts. The stronger the adoption, the more valuable recovery capabilities become.

The warning also lands against a broader market backdrop in which AI enthusiasm remains intense. Nvidia shares closed at $221.97 on Oct. 21, above both the 50-day moving average of $222.91 and the 200-day moving average of $203.12, while Adalytica’s NVIDIA earnings sentiment reads 93, or “Extreme Greed,” showing investors remain focused on AI buildout even as risk management lags behind.

Microsoft, another major AI beneficiary, closed at $532.95 on Oct. 21, above its 50-day and 200-day averages, and its filings have warned that AI products can produce unintended consequences or be used in unforeseen ways. That makes resilience a board-level issue, not just an IT function, as the next phase of AI spending shifts from model deployment to control, traceability and recovery.

Cohesity said best practice is emerging through collaboration among AI developers, security providers and governments, including initiatives such as Anthropic’s Project Glasswing and OpenAI’s Daybreak, but local standards will also matter. For UAE organisations, the immediate test is whether they can identify what an agent touched, restore clean data and credentials, and bring critical applications back in sequence after an unexpected action.

The next catalyst is not another proof point that AI can automate more work — it is whether enterprises can prove they can unwind the damage when it goes wrong.