Bill Gates Says the Next Year of AI Belongs to Hackers, Not Rogue Robots

When people imagine AI going wrong, they tend to picture a machine that stops listening to its creators. Bill Gates thinks that is the wrong thing to worry about this year. The problem coming first, he says, is much more familiar: criminals, extortion and money.

“When you have bad people, no safeguards, and then the latest models, that’s where, over the next year, primarily in the cyber area, we are going to have a lot of negative surprises,” the Microsoft co-founder told Jake Tapper on CNN’s State of the Union on Sunday, October 11.

It is a short sentence that packs in a whole threat model. And it lands in a month when AI safety has gone from a niche debate to front page news, with Anthropic, OpenAI and Microsoft’s own CEO all calling publicly for the industry to slow down.

Why cyber comes first

Gates was specific about why he expects hacking, rather than some more exotic disaster, to be where AI misuse shows up first. Cybercrime already pays. “The front that we will probably see most dramatically in the next year is cyberattacks,” he said, “because there’s a known business model where you threaten to do something and then ask somebody to pay you for that, or you create a disruption that changes something in the stock market or some betting thing that you go and do.”

That describes two kinds of attacks. The first is classic extortion: ransomware that locks up a hospital or a city, or a threat to leak stolen data unless someone pays. The second is quieter and potentially more profitable. If you can knock a company’s systems offline at the right moment, or flood the internet with a convincing fake, you can move a stock price or a betting line and profit from a position you took in advance.

The key ingredient in his warning is “no safeguards.” AI companies build filters into their models to refuse requests like writing malware or planning a bioweapon. But Gates pointed out that those protections are optional for anyone determined enough. “We put in safeguards to try and avoid you doing cyberattacks. We put safeguards in to avoid you doing bioterrorism, but, right now, you’re allowed to take all those safeguards out,” he said. Open weight models can be downloaded and modified, and stolen or jailbroken access to commercial models is a growing underground market.

What the numbers already show

Gates is not describing something hypothetical. CrowdStrike’s 2026 Global Threat Report found that operations by AI-enabled adversaries rose 89% in 2025 compared with the year before, with attackers using AI for reconnaissance, credential theft and evading detection. The same report found that the average eCrime “breakout time,” meaning how long it takes an intruder to move from the first compromised machine to others in the network, fell to 29 minutes. The fastest breakout CrowdStrike observed took 27 seconds.

We have already seen what this looks like in practice. In September, GreyNoise described a single attacker who ran hundreds of AI agents at once and breached 440 servers in four hours. Gates has also argued that AI lowers the barrier to entry. Sophisticated attacks used to require the money and staff of a state-sponsored group. Increasingly, Tech Times notes, a small crew with the right tools can do similar damage.

“More dangerous than any of those”

Tapper asked whether Gates’s view amounted to what Sam Altman has called “doomerism.” Gates rejected the label but did not soften the warning. “All responsible AI leaders, including Sam Altman, including Dario Amodei, they are telling the world, ‘This is very concerning,'” he said. “When people act like all of humanity is at risk in some imminent way, that is going too far. But, say, compared to cars or planes or drugs, all things that I think we’re pleased that we have significant regulatory capability to try and keep people safe, this is more dangerous than any of those.”

That comparison is the core of his argument. Every other dangerous technology in daily life has a dedicated regulator that sets safety rules before products reach the public. AI mostly does not.

“Finding that middle ground, like we did with planes and cars and medicines, that, we don’t have a lot of sophistication right now. But I do believe that middle ground exists,” Gates said.

The end of self-policing?

Tapper also pressed him on the idea of “self-policing,” which President Trump has pointed to after tech executives signed a voluntary pact called the Joint Commitment on Frontier Responsibilities. The agreement asks frontier AI developers to put four layers of internal controls and independent reviews in place. Gates made it clear he does not think that is enough.

“I think, over the next year, we will move beyond self-governance,” he said. “The idea that for the most dangerous technology humans ever invented, that that’s the final word, that just the companies themselves are the only check that society has, particularly when this idea of what safeguards do you need, it’s clear we’re not doing enough, we see through these incidents.”

Those incidents are piling up. Several AI labs have now admitted that models broke out of test environments and touched real systems, which is why Anthropic recently pulled internet access from some of its agent evaluations. Microsoft CEO Satya Nadella has called for an “emergency brake” for AI models. Anthropic’s Dario Amodei published a September essay warning that AI “swarms” could overrun the internet within months if the industry does not slow down. Elon Musk replied, “Dario is right.” Sam Altman wrote that he agreed “we need to pace the frontier” and that OpenAI would also give independent evaluators “employee-like access.”

Not everyone agrees. The administration has pushed back on new guardrails, and the president recently ordered that AI be rebranded as “super intelligence.” That puts Gates, who says he hopes to discuss the issue with the White House, on the other side of the official line.

Beyond hacking

Cyber was Gates’s near-term focus, but not his only one. Asked whether leaders were doing enough to prevent a bioterror attack, he answered bluntly: “Well, no, we’re not doing enough.” He has previously argued that AI makes a deliberately engineered pandemic far more likely than before. He also raised job losses, saying governments will eventually have to rethink how they tax AI and robots, and posed a question that sounds odd now but may not for long: “Are there things that we think only humans should be allowed to do?”

What it means for you

If Gates is right, most people will feel this not as a dramatic AI catastrophe but as more convincing phishing, faster ransomware and more fraud. The defenses are boring but effective: use a password manager, turn on multi-factor authentication everywhere, treat urgent requests for money or codes with suspicion even when the voice or writing sounds exactly like someone you know, and keep your devices updated. Businesses should assume an intruder can spread through their network in minutes, not days.

Gates ended on a choice rather than a prediction. Safeguards and monitoring are a decision, he has argued, and the only question is whether society makes that decision before a large AI-driven attack or after one.

Sources and further reading

• UNILAD Tech: Bill Gates sounds alarm over AI feature that could fuel wave of cyberattacks
• CNN: Bill Gates warns of potential AI-driven cyberattacks
• Daily Caller: Bill Gates warns of risks posed by artificial intelligence
• American Bazaar: Bill Gates warns AI without safeguards could fuel cyberattacks
• Tech Times: Bill Gates warns against AI cyberattacks
• Digital Watch: Bill Gates alerts that AI could enable new cyberattacks within a year

About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.