The policy response to AI cybersecurity threats

With a spate of ‘rogue agent’ hacks, the cybersecurity threats posed by frontier artificial intelligence models have moved from theoretical to a clear and present danger. So far, these hacks have primarily been perpetrated accidentally, either due to frontier models breaking containment, or loosened controls for capability testing. Nevertheless, they can have real-world consequences.

As AI proliferates, supervisory functions must evolve to mitigate the risks it poses. That will require the establishment of best practices on AI monitoring, designation of supervisors, AI labs’ obligations to them and their powers of enforcement.

OMFIF is hosting a roundtable to explore these topics with experts from financial and technology supervisors, and those working in the AI industry.