Secure KI requires more than research

We need a security institute for KI, we wrote in DN on February 21. Sven Størmer Thaulow and Jon Atle Gulla of NorWai point out, in their reply, that Norway needs to develop AI models, not just test them.

We share that vision. But a commitment to security and development is not a contradiction. Just look to the US and UK, which combine national security institutes with ambitious KI development. The security institutes build the trust that makes use possible.

Much of the testing will have to take place locally where the models are developed. But a state security body can both act as a competence center for secure KI, and coordinate external testing when needed. This can be taken care of by a small and agile organisation in KI-Norway, which draws on external expertise where needed.

KI is about to sail up as the biggest security challenge we face. In such a situation, we need to build closer alliances with our close allies. The international network of AI Safety Institutes is a perfect opportunity for this.

We must also increase the competence of the political leadership on IT security, which will have to make strategic decisions in the time ahead in order to secure Norway. A state KI security body can act as a bridge builder between the technical and diplomatic environments and provide up-to-date analyses to political leadership.

No research community or company, no matter how adept they are, can play these features.