NSA, CISA, FBI and allied agencies publish Deploying AI Systems Securely

On 15 April 2024 the Artificial Intelligence Security Center at the National Security Agency (NSA AISC) published the joint Cybersecurity Information Sheet Deploying AI Systems Securely. It was produced in collaboration with CISA, the FBI and national cyber agencies of Australia, Canada, New Zealand and the United Kingdom.

What it is

The sheet is written for organizations that deploy and operate AI systems built by someone else, rather than for the teams that build models. It sets three aims.

• Improve the confidentiality, integrity and availability of AI systems.

• Make sure known vulnerabilities in AI systems have appropriate mitigations.

• Provide methods and controls to protect, detect and respond to malicious activity against AI systems and their related data and services.

What it requires

Nothing is compulsory. CISA encouraged organizations that deploy and operate externally developed AI to review the guidance and apply it as applicable, and pointed them to earlier joint publications on securing AI, including Guidelines for secure AI system development.

What security leaders should do

• Put every third-party model, hosted AI service and embedded AI feature in scope; this sheet speaks directly to organizations in the deployer position.
• Score each AI deployment against the three aims: confidentiality, integrity and availability; mitigation of known vulnerabilities; and detection and response coverage.
• Extend logging and incident response playbooks to AI components and to the data stores they read and write.
• Use the sheet alongside the earlier secure development guidelines in supplier negotiations, so build-side and deploy-side expectations line up.