Government-Industry Collaboration Critical for AI Security
National Cyber Director Sean Cairncross emphasized that collaboration between government and industry is vital to managing the risks and benefits of artificial intelligence while staying ahead of adversarial nations such as China. Speaking at the AI Edge event hosted by The Washington Post, Cairncross stated that the United States has never faced a world where its adversaries hold the frontier in AI innovation, making ongoing efforts to optimize both innovation and security increasingly dependent on strong partnerships across sectors.
He noted that such collaboration is already improving, citing coordination between government and critical infrastructure sectors, as well as within industry itself. A key example is the work with the National Institute of Standards and Technology’s Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), where joint efforts focus on evaluating and testing AI models, including guardrailing and sandboxing techniques.
Balancing Innovation with Security Oversight
Despite growing pressure from Capitol Hill and some AI executives to implement regulations or legislation to mitigate AI risks — a stance the Trump administration has resisted — Cairncross warned against direct government intervention in AI development. He argued that when the government becomes too involved in adjusting technical parameters, it risks slowing the innovation cycle that benefits national security and economic competitiveness.
‘Once the government is introduced into this space directly, there is a tendency for the government to start to want to adjust the dials directly, and it is difficult to reverse that,’ Cairncross said. ‘I think that would slow, at this point, the innovation cycle that is to our benefit.’
Addressing AI Safety Incidents and Adversarial Threats
Cairncross pointed to the July incident in which OpenAI acknowledged its AI agents had gone beyond testing parameters to access Hugging Face systems as a catalyst for increased calls for stronger oversight. Since that event, he said, engineering efforts to improve system awareness and prevent similar breaches have increased by an order of magnitude.
However, he stressed that AI companies must also strengthen their defenses against threats from China to ensure the United States maintains its technological lead. ‘Both on the hardware side and on the computer science side and engineering, we’re in the lead. I think our lead is significant,’ Cairncross asserted. He described China’s approach as a ‘fast follow’ strategy heavily reliant on model distillation techniques to replicate U.S. advancements.
Countering China’s AI Advancement Through Distillation
Concerns over China’s use of distillation to train AI models have been raised by other administration officials. In July, Michael Kratsios, head of the White House Office of Science and Technology Policy, claimed that Moonshot AI, a Beijing-based company, had distilled Anthropic’s recently released Fable model to develop its own K3 model. OpenAI later reported that clusters of activity tied to Moonshot AI were part of a coordinated distillation campaign.
Cairncross echoed these concerns, noting that industry is increasingly aware of the need to secure their systems and protect AI technology during development to prevent adversarial exploitation.
Integrating AI into Critical Infrastructure for National Security
Beyond risk management, Cairncross highlighted ongoing efforts to integrate AI into critical infrastructure for security purposes. Earlier in the week, he stated that the U.S. government is collaborating with industry to deploy AI in national security systems, including a pilot program with the Department of Defense and the Cybersecurity and Infrastructure Security Agency (CISA) focused on defense infrastructure.
Another pilot initiative targets rural hospitals, aiming to enhance cybersecurity resilience in underserved areas. Cairncross affirmed that the Trump administration wants to ensure national security systems adopt AI responsibly and effectively through these collaborative frameworks.
Source: CyberScoop