The Pulse

Lava Security's neocloud and AI data center risk list joins OWASP as an incubator project

Lava Security's framework for assessing neocloud and AI data center risks is now an OWASP incubator project. The guidance covers hardware, networking, tenant isolation and other infrastructure layers.

AI.info Team · · fact-checked · against 7 sources ↓ · Published 11 Nov 2026

Lava Security's framework for identifying risks in neoclouds and AI data centers has become an official OWASP project. OWASP lists it as an incubator project for security guidance covering the infrastructure on which AI workloads run.

The framework was initially published in July 2026. The project's GitHub issue tracker contains issues opened on October 5, 2026.

Beyond the model

The framework was developed by Yakir Kadkoda, Lava's chief technology officer and co-founder, and Michael Katchinskiy, its head of security research. OWASP identifies both as project co-leads.

More than 20 practitioners across neocloud providers, hardware manufacturers, security companies, universities and research institutions contributed feedback during development. The work also drew on disclosed vulnerabilities, incident reports, technical documentation and Lava's research.

The guidance covers physical hardware, networking, management and orchestration systems, storage and the operational environments hosting workloads. Its scope does not center on AI models or application-layer problems such as prompt injection and insecure agent behavior.

The intended audience includes infrastructure providers and operators, customers buying cloud capacity and security teams. Each risk entry contains a description, potential impacts and failure modes, attack scenarios, references and mitigation strategies for providers and customers.

Risks below the operating system

Among the scenarios in the release is the reassignment of a bare-metal server after a customer has had root access. A compromised hardware boot configuration could persist through an operating-system reinstall and affect the next customer assigned that server.

The risk list is ordered by severity and includes assessments of likelihood, impact and detection difficulty. Its top four risks cover hardware and firmware integrity, network and interconnect vulnerabilities, multi-tenant isolation and resource reuse, and out-of-band management planes.

The project ranks hardware and firmware integrity compromise as critical, with medium likelihood, severe impact and high detection difficulty. It also rates unsafe multi-tenant isolation and resource reuse as critical, while assigning that risk low likelihood, severe impact and very high detection difficulty.

Other entries address supply-chain compromise, facility and data center management systems, handling of data and AI artifacts, certification gaps and provider transparency, operational infrastructure services, and vendor embargo gaps and patch velocity.

Open development

Security researchers, infrastructure engineers, hardware vendors and operators can review the guidance, challenge its assumptions and propose changes through the open development process on GitHub. The project also invites additional research and mitigation recommendations.

Contributors will be credited in future releases. The framework is intended to change as infrastructure does, including as new GPU architectures, networking technologies and deployment models bring further security considerations.