New OWASP Top 10 Exposes Hidden Security Weaknesses in Neocloud and AI Infrastructure

Blog WriterCybersecurity News - Original News Source is cybersecuritynews.com


A new OWASP Incubator Project highlights an overlooked AI security gap: the servers, networks, and data centers supporting AI systems receive far less attention than models, apps, and agents.

The OWASP Top 10 Neocloud and AI Data Center Security Risks tries to fix that. It gives neocloud providers, their customers, and AI data center operators one shared list of the biggest infrastructure risks.

AI data centers pack thousands of GPUs, fast network fabrics, remote management controllers, shared storage, and complex scheduling tools into one place.

Much of this gear came from high-performance computing, where users were trusted. Today, the same parts run commercial GPU clouds, where unrelated companies share the same hardware. A flaw once considered minor can now put many customers at risk.

Many security tools also cannot see this layer. They watch operating systems, apps, and normal networks. They often miss firmware, management ports, and the special links that join GPU nodes. If the base is compromised, nothing running on top can be trusted.

Michael Katchinskiy and Yakir Kadkoda of the security firm Lava lead the project. The team selected risks from security research, incident reports, published CVEs, vendor documents, and practitioner talks.

Each candidate was scored on likelihood, impact, and detectability. Reviewers came from companies and schools such as Dell, Nebius, Google, IBM, and the University of Basel.

OWASP Top 10 Exposes Neocloud AI Risks

The project does not cover prompt injection or model abuse. Other frameworks, including the OWASP Top 10 for LLM Applications and MITRE ATLAS, handle those.

OWASP ranks the risks from most to least severe. The first five are rated critical. These are hardware and firmware tampering, weak networks between GPUs, poor separation between tenants, unsafe out-of-band management, and supply chain compromise.

Lower on the list are weak facility systems, careless handling of data and model files, gaps in certifications and provider honesty, unsafe operations tools, and slow patching.

Firmware and management controllers deserve a close look. A baseboard management controller, or BMC, lets admins control a server even when its operating system is off. If a hacker takes it over, the access can survive a reinstall and follow the server to the next customer.

OWASP Top 10 Risks:

# Risk What can go wrong
1 Hardware and firmware integrity compromise Hidden tampering below the OS survives reinstalls and resets, so access persists after a server is reassigned
2 Network and interconnect vulnerabilities Weakly secured InfiniBand, RoCE, and RDMA links allow unauthorized access, data interception or changes, and movement between GPU nodes
3 Unsafe multi-tenant isolation and resource reuse Failed isolation or uncleaned servers expose one tenant’s data, workloads, or infrastructure to another
4 Insecure out-of-band management plane A hacked BMC can restart servers, alter firmware, open remote consoles, or take over whole fleets
5 AI infrastructure supply chain compromise Compromised hardware, firmware, drivers, OS images, or updates bring in malicious code or hidden access
6 Insecure facility and data center management systems Weak cooling, power, and access control systems can disrupt operations or leak sensitive infrastructure details
7 Insecure data and artifact handling Poor access controls or cleanup lead to stolen model weights, tampered training data, or recovered leftover files
8 Certification gaps and provider transparency failures Audits may skip the GPU layer, and undisclosed resale of capacity hides who runs the hardware
9 Insecure operational infrastructure services Compromised deployment and management tools let attackers push malicious workloads, change configs, or disable controls
10 Vendor embargo gaps and patch velocity failures Late disclosure and slow or disruptive patching give attackers a longer window to exploit known flaws

This is not just theory. Researchers recently found 36,872 internet-exposed IPMI interfaces, and about 67 percent leaked password-based authentication data before login.

Some belonged to GPU and bare-metal cloud providers. Earlier, a flaw in AMI MegaRAC software (CVE-2024-54085) let attackers skip login on affected servers.

Our own BMC firmware vulnerabilities report showed how weak Supermicro patches left the door open. Tenant isolation is another weak spot.

The NVIDIAScape container escape flaw, CVE-2025-23266, let a malicious container gain root access on a GPU host. It hit toolkit versions up to 1.17.7. That is the exact kind of problem the OWASP list warns about.

The framework notes that attackers have begun targeting AI infrastructure. It cites a TIME report about a US tech firm whose AI data center lost intellectual property.

It also cites a conviction for stealing AI data center designs and claims of a 10-petabyte theft from China’s Tianjin supercomputing center. The authors also warn that GPU shortages mean customers often pick whatever provider has capacity, not the most secure one.

Providers can use the list to review their setups and show maturity. Customers can use it when buying, writing contracts, and comparing vendors. Ask a provider whether its audit covers the GPU hardware, and whether it resells capacity from someone else.

Security teams should keep BMCs off the public internet, patch firmware, and put management traffic on its own network. Our AI data center security checklist offers a 60-day plan to do that.

Give your SOC the alert context on the latest threats to stop incidents quickly -> Integrate Threat Intel from 16K SOCs

The post New OWASP Top 10 Exposes Hidden Security Weaknesses in Neocloud and AI Infrastructure appeared first on Cyber Security News.