Bill Gates warns cyberattacks are where the negative surprises will come over the next year, and rejects letting AI companies regulate themselves

What he actually said. AFP's October 11 report captures Gates in a Sunday interview with ABC News. On the order of risks, his words were that the front we will probably see most dramatically in the next year is cyberattacks. He then separated the timescales very explicitly: the real danger right now is not AI going out of control — work on that is still needed, but in terms of big impact it is years away. He assigned the near-term risk to people: what is really bad is bad people using AI, particularly in settings where you can take all the safeguards out. Then comes the bluntest line: right now, you are allowed to take all those safeguards out. So when you have bad people, no safeguards, and the latest models, that is where, over the next year, primarily in the cyber area, we are going to have a lot of negative surprises.

What he is arguing against. In the same interview, Gates dismissed the proposition that self-regulation by AI companies is sufficient to keep the world safe. AFP's report directly identifies that position as one advocated by US President Donald Trump. Gates' words: the idea that for the most dangerous technology humans ever invented, that is the final word, that just the companies themselves are the only check that society has — it is clear we are not doing enough. What he argues for is increased restrictions, aimed at preventing bad actors from using AI for bioterrorism and cyberattacks. AFP places him in a sequence: he is the latest tech magnate to sound the alarm on the threats posed by the new technology. That framing alone shows how much the conversation about AI risk has heated up among industry leaders.

This is not a sudden change of position. AFP notes the background: Gates published a long essay in August warning that the world is dangerously unprepared for the upheaval AI will bring, especially to employment. The original — The choices we make about AI now are critical — is on Gates Notes. In it he writes that AI capabilities are starting to be used for cyberattacks, and that the smartest cybersecurity experts he knows are scared about the next few years because attackers are getting powerful new capabilities faster than defenders can fix all the weaknesses. He also states the key symmetry: the same AI model that can find a flaw in software so a company can fix it can also help a criminal exploit it. That sentence captures the hardest part of AI security — offence and defence draw on the same capabilities, and defenders have structurally less room for error.

Three things are worth noticing. First, separating risks by timescale is itself a policy judgment. Placing AI going out of control years away and bad people using AI within the next twelve months means that, in his view, visible resources should go first to present-day misuse rather than distant existential risk. Second, his regulatory stance is not in tune with some of the prevailing industry language. He characterises company self-regulation as insufficient to be the final answer and stresses that society needs a check outside the companies — related to current debates about voluntary commitments, transparency reporting and safeguards evaluation, but a harder demand. Third, the specific risk categories he names are cyberattacks and bioterrorism, both directions where capability can be misused directly and consequences are hard to reverse. Overall, the value of the interview is that it offers a fairly concrete list of risks and an explicit timetable, rather than a generic warning that AI is risky.