The hidden risks of cloud AI are a security crisis in the making

Within the promise of AI-driven efficiency lies an unsettling reality: cloud-based AI systems are riddled with security pitfalls that could expose businesses to significant cyberthreats.

The inherent vulnerabilities of AI in the cloud stem from a dangerous mix of misconfigurations, excessive permissions and poor security hygiene. AI services and frameworks are often left publicly exposed, with storage buckets that lack encryption, over-privileged identities that grant unnecessary access, and minimal auditing to track malicious activity. These gaps create a perfect storm for cybercriminals seeking to manipulate AI models, extract sensitive data and undermine the integrity of AI-generated outputs.

Recent data paints a grim picture: approximately 70% of cloud AI workloads contain at least one unremediated vulnerability. Worse still, Tenable’s 2025 Cloud AI Risk Report identified CVE-2023-38545 — a critical curl vulnerability — in 30% of cloud AI workloads. This is not a minor oversight; it is a glaring weakness that invites attackers to infiltrate, manipulate and exploit AI systems with ease.

Threat actors are increasingly targeting AI models not just to steal data but to corrupt the very algorithms businesses rely on. By poisoning training and testing datasets, attackers can influence AI decision-making, generating biased or harmful results that damage businesses and erode customer trust. Worse still, AI models are often trained on datasets that contain sensitive intellectual property, personal information (PI) or even classified corporate data. When this information is inadequately protected, it becomes a goldmine for exploitation.

The consequences of compromised AI models are far-reaching. Imagine an AI-powered fraud detection system manipulated to overlook suspicious transactions, or an AI-driven recruitment tool tricked into making biased hiring decisions. These aren't just hypothetical risks; they are real-world scenarios that could cripple business operations and trigger regulatory scrutiny.

A reactive approach to AI security is no longer viable. As AI adoption surges, organisations must implement a mature exposure management strategy that proactively addresses AI-related risks. This requires:

• Managing AI exposure holistically: AI models, datasets and cloud environments must be continuously monitored for vulnerabilities. Security teams must integrate AI-related risks into their broader cloud security posture, ensuring unified visibility across infrastructure, identities and workloads.
• Classifying AI assets as high risk: All AI components, especially those linked to sensitive business assets, should be classified as critical. AI data — including test datasets — must be treated with the same security rigor as production environments to prevent unauthorised leaks.
• Tightening identity and access controls: Over-privileged identities remain a glaring weakness in cloud AI environments. Businesses must enforce the principle of least privilege, ensuring that only essential personnel have access to AI models and datasets. Role-based access controls and continuous monitoring tools should be standard practice.
• Prioritising AI-specific threat remediation: Not all vulnerabilities pose the same level of risk. Organisations need advanced security solutions that prioritise AI-related risks based on business impact, filtering out unnecessary noise and focusing on the most critical threats.
   

Cloud-based AI is a double-edged sword. While it unlocks innovation and efficiency, its security risks cannot be ignored. AI-powered cyber attacks are no longer a question of ‘if’ but ‘when’, and businesses that fail to secure their AI assets will find themselves exposed to operational disruptions, data breaches and reputational damage.

Security leaders must act now to close AI security gaps before they become existential threats. The future of AI depends not just on its capabilities, but on the safeguards we put in place to protect it.

Image credit: iStock.com/Olemedia