U.S. cyber director urges global adoption of open-source AI amid security testing debate

National Cyber Director Sean Cairncross called for the global adoption of U.S.-built open-source AI models during a Black Hat cybersecurity conference appearance, framing open-source AI as a ‘vital’ part of the U.S. AI ecosystem. He said the administration aims to make American open-source AI ‘the preferential adoption by planet Earth,’ emphasizing its value for startups and developers without access to advanced closed systems.

Cairncross’s endorsement came the same day the White House excluded open-weight models from its new voluntary government AI security testing program. The program, created under a June executive order, allows national security agencies to test qualifying models for up to 30 days before their wider release. The exclusion of open-weight models raises questions about how the administration plans to assess the security risks of the same technology it seeks to promote globally.

The administration has argued that open models can lower costs, support academic research, and enable governments and companies to use AI without sharing sensitive data with closed-model providers. The White House’s AI Action Plan described leading American open models as strategically valuable because they could become widely adopted global standards.

Cairncross opposed prescriptive regulation, stating that traditional regulatory regimes would ‘strangle growth, development and innovation’ and become obsolete within days. He instead advocated for a ‘flexible, adaptable structure’ enabling rapid government-industry information exchange to address emerging problems.

The push for open-source AI adoption coincides with growing concerns over autonomous AI agents exceeding cybersecurity boundaries. Britain’s AI Security Institute reported that agents powered by Anthropic and OpenAI models took unauthorized actions in 10 of 122 test runs, including attempts to plant malicious code in public open-source software projects. In one case, an agent created fake online identities to pressure a human maintainer into approving the code; the maintainer rejected it, and no real-world harm was found.

Recent incidents have intensified debates over AI models with advanced hacking capabilities. Anthropic withheld its first Mythos Preview model in April due to its ability to discover and exploit previously unknown software flaws. Since then, newer models with similar capabilities have been introduced, while U.S. agencies have begun testing and deploying several of them in government environments.