Why Global AI Security Cooperation Matters More Now

AI security is now an international problem. The same models, tools and attack techniques cross borders in days, and as AI moves deeper into software, operations and decision support, a weakness in one model can create risks no single company or country can handle alone.

That is why joint government guidance matters. It moves the conversation past hype toward practical safeguards: protecting training data, testing models, assessing risk and setting clear expectations for developers. Coordination won’t solve the problem by itself. It can raise the floor.

The guidance itself

On May 22, 2025, security agencies from the United States, the United Kingdom, Australia and New Zealand released joint guidance titled “AI Data Security: Best Practices for Securing Data Used to Train & Operate AI Systems.” It is aimed at organizations that build or run AI systems.

What the guidance asks for

The document covers risks across the AI lifecycle, from development and testing to deployment and operation. It singles out three data risks: weaknesses in the data supply chain, maliciously modified or “poisoned” data, and data drift. Recommended controls include encryption, digital signatures, tracking where data comes from and secure storage on trusted infrastructure.

Who wrote it

The guidance came from the NSA’s Artificial Intelligence Security Center (AISC), CISA and the FBI in the United States, together with Australia’s ACSC, New Zealand’s NCSC and the UK’s NCSC. Their names carry weight with the companies and agencies that follow their advice.

Where the focus falls

The agencies stress ongoing risk assessments, since data used by an AI system can change or be tampered with after deployment. They also want organizations to share what they learn, because attackers reuse techniques across countries.

Learning from past incidents

AI systems have already shown they can be manipulated through bad data. The guidance builds on those lessons and asks developers to plan for attacks on the data their models depend on, not only on the code.

A model for future cooperation

The joint release shows allied agencies can agree on practical AI security advice. Expect more of it as AI spreads into critical systems.

What organizations should do

The guidance from the United States, the United Kingdom, Australia and New Zealand gives organizations a concrete checklist. Teams building or buying AI systems should review it, map where their training and operating data comes from and add controls such as signing and provenance tracking where they are missing.

Related hub: For a fuller view of AI governance, security risk, and what practical adoption should look like, continue to AI in Cybersecurity in 2026.

Frank Jones, CISSP

Frank Jones has loved computers from the age of 13. Frank got his hacking career started when he downloaded a war dialing program that he used to detect dial up modems in his hometown of Chicago. Frank Jones now works as a JAVA coder and cyber security researcher.