Bill Gates Warns AI Is in a Period of Extreme Vulnerability to Cyberattacks

In an interview published on September 29, Bill Gates told New York Times podcast host Ezra Klein that the ability to carry out cyberattacks capable of scrambling bank accounts and shutting down the electric grid already exists today, and that AI is in a period of extreme vulnerability.

Gates recalled Anthropic's Mythos model announced in April, saying the hacking capability demonstrated by such models was stunning. He also referenced Anthropic's Project Glasswing, which proposed giving AI to a small group of people to fix bugs before widespread use, but noted there is too much code for that to work effectively. He argued that the danger from open-source models does not come from AI gaining sentience or turning evil, but from humans who may use these models with bad intentions. Gates advocated for a supervisory layer with absolutes across all models.

He pointed to aspects of Hugging Face as evidence that models need safeguards and monitoring. Gates described AI development as "like evolutionary history" and said "the aliens really are here and have come," but his most acute concern is that present-day models are already out in the world.