Securing the AI Lifecycle: Deploy

The deploy phase transitions the model from controlled development to real-world use, exposing it to new security and operational risks. Integration with infrastructure and external networks makes deployed models prime targets for exploitation. Securing this stage through hardened APIs, strict access controls, and isolated environments is essential for preserving model integrity and protecting downstream systems.

Model Weights

Model weights (also parameters, learned representations, or control policies) constitute the core intellectual property and operational logic of AI systems that employ them. If stolen or altered, they can lead to model replication, performance degradation, or targeted attacks. Protecting weights with encryption, access controls, and version monitoring helps ensure the integrity and security of deployed models.

Subcategory	Identifier	Action or Control	OWASP and BIML Threats Mitigated	Why the Control Is Relevant
Access control and authorization	MW.ACA-2	Restrict access to model weights using role-based and privilege-based controls	API5:2023 — Broken Function Level Authorization LLM10: Model Theft ML05: Model Stealing Attack	Restricting access to model weights using role‑ and privilege‑based controls helps reduce the likelihood of unauthorized access, theft, or manipulation by ensuring that only properly authorized personnel can interact with sensitive model assets.
	MW.ACA-1	Encrypt model weights and restrict access to authorized personnel only	LLM10: Model Theft ML05: Model Stealing Attack BIML: Model Extraction	Encrypting model weights and limiting access to authorized personnel helps reduce the risk of unauthorized retrieval, replication, or misuse of proprietary AI models.
Storage security and isolation	MW.SSI-1	Store weights on secure servers, not local devices	LLM10: Model Theft ML05: Model Stealing Attack BIML: Model Extraction	Storing model weights on secure servers rather than local devices helps reduce exposure to unauthorized access or extraction, lowering the risk of theft or tampering.
Cryptographic protection of model weights	MW.CPMW-1	Encrypt model weights at rest and in transit to ensure confidentiality and integrity	LLM10: Model Theft ML05: Model Stealing Attack	Encrypting model weights at rest and in transit helps maintain their confidentiality and integrity, reducing the likelihood of unauthorized access or extraction during storage or transmission.
	MW.CPMW-2	Use cryptographic techniques to protect model weights during generation and storage	LLM10: Model Theft ML05: Model Stealing Attack BIML: Model Extraction	Using cryptographic techniques during model weight generation and storage helps lower the risk of unauthorized access or extraction by ensuring that the weights remain protected throughout their lifecycle.
	MW.CPMW-3	Implement hardware-based encryption to secure model weights at rest	LLM10: Model Theft ML05: Model Stealing Attack	Implementing hardware‑based encryption helps strengthen protection of model weights at rest, reducing the likelihood of unauthorized access, theft, or tampering.