Darktrace's 2026 State of AI Cybersecurity report, based on a survey of more than 1,500 security professionals globally, delivers findings that should redefine how UAE security leaders brief their boards and structure their defense programs. Eighty-seven percent of respondents report that AI is significantly increasing malware sophistication and success rates. The same percentage report that AI is significantly increasing their security operations workload. Seventy-three percent say AI-powered threats are having a significant impact on their organisation right now, and ninety-two percent say those threats are forcing them to upgrade their defenses. For UAE businesses navigating the CBUAE Cyber Risk Management Framework, DIFC data protection obligations, and ADGM regulatory requirements, these are not abstract statistics. They are an empirical description of the environment in which compliance programs must operate.

Sixty-two percent of respondents in the Darktrace survey have experienced a deepfake social engineering attack. This is not a future risk category. It is a current operational reality for the majority of surveyed organisations, and AI-enabled ransomware incidents rose over 30% in 2025. Given the UAE's role as a regional financial and commercial hub, the prevalence among UAE-specifically targeted organisations is likely at least comparable to the global average.
Thirty-two percent have experienced a prompt injection attack, a category unique to organisations that have deployed AI tools internally. In a prompt injection attack, an adversary constructs inputs designed to manipulate AI system behaviour, causing the system to take actions its operators did not authorise. As UAE financial services firms, real estate platforms, and government entities accelerate AI tool deployment under Vision 2031 objectives, the attack surface for prompt injection expands proportionally.

The top three AI-enabled threats that security leaders report fearing most are hyper-personalised phishing, cited by fifty percent of respondents, automated vulnerability scanning at forty-five percent, and deepfake voice fraud at forty percent, with ransomware-as-a-service platforms such as Lockbit3 showing how AI can adapt attack sequences in real time to exploit vulnerabilities with greater efficiency. Each of these threat categories directly targets human decision-making rather than technical controls alone, which means that technical defense investment without corresponding human awareness and response capability leaves a significant portion of the threat surface unaddressed in a convincing manner for many organisations to detect early. Autonomous malware can remain dormant inside enterprise networks for weeks, monitoring activity before attackers move. This also reflects how identity and credential abuse have now surpassed traditional network perimeters as primary attack vectors, making behavioral AI crucial for helping teams detect unusual activity earlier and prevent breaches across distributed environments.
Forty-six percent of security leaders say they feel inadequately prepared to address AI-powered threats, a figure that has risen from forty-five percent in 2025. The preparation gap is not narrowing. It is widening, even as investment in security defense increases. The Darktrace report identifies the primary barrier as insufficient knowledge or use of AI-driven countermeasures, not budget and not headcount. Only 37% of organisations globally maintain formal AI usage policies. This is a significant finding because it implies that the constraint is knowledge and strategic direction rather than resources, which means that organisations willing to invest in understanding before spending have an opportunity to close the gap faster than those who increase spending without strategic clarity; it also underscores the importance of consistent governance, since cybersecurity effectiveness can be compromised by inconsistent AI governance and oversight.

Ninety-two percent of respondents say AI threats are forcing them to upgrade their defenses. The gap between the ninety-two percent who recognise the need and the fifty-four percent who feel adequately prepared represents the population of organisations that have acknowledged the problem but have not yet translated that acknowledgement into effective capability. Separately, 92% of organisations express concerns over autonomous AI agents in workplaces, especially when actions outside expected parameters are not flagged for human review. For UAE security leaders, this gap also reflects the distance between acquiring AI tools and establishing trust in their effective integration, which sharpens both the risk of underpreparedness and the opportunity to move into the prepared minority before a major AI-driven incident occurs and differentiate faster than peers.
The CBUAE Cyber Risk Management Framework requires financial institutions to maintain risk assessments that account for emerging and evolving threat vectors, and the UAE Cyber Security Council has also promoted zero-trust frameworks. The Darktrace report provides empirical evidence that AI-powered attacks are no longer emerging, they are current and prevalent across the financial sector and the wider world. In 2023, UAE banks established cooperation with the US Treasury on cybersecurity, reinforcing the international dimension of financial-sector defense. An institution whose risk register does not explicitly address AI-powered threat categories including deepfake social engineering and automated vulnerability scanning is operating with a risk assessment that is demonstrably out of date.
DIFC and ADGM data protection obligations extend to fraudulent impersonation of clients and staff. Deepfake social engineering attacks that manipulate employees into disclosing information, approving transactions, or granting access represent a direct data protection risk under these frameworks, not merely a cybersecurity risk. If these incidents lead to data exfiltration involving personal data, PDPL breach notification requirements may apply. Sixty-two percent of surveyed organisations have experienced such attacks. An organisation that has not trained its staff to recognise and respond to deepfake social engineering is not meeting its reasonable security obligations under DIFC or ADGM rules.
UAE Vision 2031 positions Dubai and the wider Middle East for leadership in artificial intelligence, and the UAE’s AI industry is projected to reach $23.5 billion by 2030. Greater AI adoption means greater AI attack surface. Each new AI system deployed creates potential prompt injection targets, data processing risks, and new social engineering attack vectors. The UAE Government has also emphasized developing cyber security talent in AI, including comments associated with Mohamed Al Kuwaiti. Security programs that were designed before AI became central to operations need to be reassessed against the threat model that AI adoption creates, not merely expanded with additional budget.
The Darktrace report's identification of insufficient knowledge as the primary barrier suggests that the highest-leverage first investment is education and strategic alignment, not technology procurement. Board-level briefings that translate AI threat statistics into business risk language framing deep-fake fraud in terms of financial exposure and regulatory consequence rather than technical vulnerability classes create the mandate for subsequent investment decisions. AI tools should also be governed as privileged digital actors within the organisation.
Deepfake social engineering awareness training specific to UAE financial services contexts is a direct response to the sixty-two percent prevalence figure. Training that focuses on the scenarios most relevant to UAE organizations impersonation of executives, clients, and counterparties in real estate transactions, banking approvals, and regulatory communications is more effective than generic awareness content. It should also cover stronger authentication measures and more innovative user training to combat AI-driven phishing in a timely manner.
AI Security Architecture Reviews that assess current AI tool deployments against the prompt injection risk surface give security teams a concrete inventory of exposure before an attack demonstrates it. Organisations should also enhance visibility into AI usage and control access to sensitive data as part of the review scope. For organisations deploying AI in customer-facing, transaction-processing, or regulatory reporting contexts, this review is a compliance-relevant exercise under CBUAE and DIFC frameworks, not only a security best practice.
SHELT's advisory services help UAE security teams and boards develop the knowledge base and strategic posture needed to address AI-powered threats within the UAE regulatory framework, including enterprise use of secure AI agents in regulated deployments. Our AI Threat Landscape Briefings translate the Darktrace report's findings and similar research into specific board-level risk language calibrated to UAE financial services, real estate, and government contexts where AI agents may handle sensitive workflows.
Our Deepfake and Social Engineering Awareness Training programme is built specifically for UAE environments, using scenarios grounded in the transaction types, communication channels, and regulatory contexts that UAE financial services and commercial organisations operate in. Staff who can recognise a deepfake impersonation attempt in a wire transfer approval context are not interchangeable with staff who have completed generic phishing awareness training.
For organisations at the strategic level assessing whether their current security architecture is positioned for the AI threat environment that the Darktrace report describes SHELT provides AI Security Architecture Reviews that produces a structured gap analysis against CBUAE, DIFC, and ADGM requirements with a prioritised remediation roadmap. The review should cover defined security requirements before development, tool permissions, scope limits, and controls to implement layered protections. It should also include testing for prompt-injection weaknesses and ways to secure customer-facing or support workflows that ingest external content.
Investment in AI security tooling addresses a portion of the AI threat surface, and organisations often invest expecting stronger operational efficiency, but the Darktrace report specifically identifies insufficient knowledge of AI-driven countermeasures, not insufficient tooling, as the primary preparation barrier. Technology investment without strategic clarity about which threat categories are being addressed, how the tools are being operated, and whether actions outside expected parameters are escalated for human review can create a false sense of readiness, and concerns about autonomous agents remain common even where organisations have invested heavily. The preparation assessment that matters is not whether the tools are in place, but whether the organisation has mapped its AI threat surface, trained its staff on the human-facing attack vectors that tools cannot fully address, reviewed its risk register to reflect the AI threat categories now documented as prevalent in the Darktrace research, and established visibility into AI usage and control over access to sensitive data.
Prompt injection is an attack technique that manipulates AI systems by crafting inputs designed to override or subvert the system's intended behaviour, and indirect prompt injection can be more dangerous than direct prompt injection when external content is ingested. In a financial services context, the risk depends on how AI agents are used in UAE fintech and banking environments, where they often have tool access to payment systems. If AI tools are involved in customer communication, document processing, transaction screening, or regulatory reporting, a prompt injection attack could cause the AI to produce incorrect outputs, approve transactions it should decline, or generate regulatory documentation that does not reflect actual transaction data; if an agent is compromised, it could also trigger unauthorized transactions or data exfiltration, and PDPL breach notification requirements may apply. Customer support tickets or other external content can carry hidden instructions, so reviews should cover those workflows as well. The thirty-two percent prevalence in the Darktrace survey, one in three organizations having experienced such an attack, should prompt any UAE financial institution that has deployed AI in operational roles to review those deployments specifically for prompt injection exposure. This review is a reasonable security practice under CBUAE Cyber Risk Management Framework expectations.
The most effective board briefings on AI threats translate the technical findings into financial exposure and regulatory consequence language. The Darktrace statistics 87% prevalence of AI-increased attack sophistication, 62% deepfake attack experience provide the empirical grounding for a risk discussion framed around questions boards can act on: What is the potential financial exposure from a successful deepfake fraud event in our transaction approval processes? What are the regulatory reporting requirements if a deepfake social engineering attack results in a data breach? Does our current security program explicitly address AI-powered threat categories in our risk register and incident response playbooks? Do we have enough visibility into AI usage across the organisation, and is access to sensitive information appropriately controlled? SHELT's board briefing service prepares these materials in formats calibrated to UAE financial services governance expectations, with specific reference to CBUAE, DIFC, and ADGM regulatory contexts, and helps leadership discuss next-step security requirements.
.png)
© SHELT 2023 Privacy Policy | Terms & Conditions