AI Could Be Used to Attack Power Grids, Hospitals, and Water Systems

Some of the biggest names in technology are sounding the alarm about a serious danger. Companies like OpenAI, Google, Microsoft, Anthropic, and Oracle signed an open letter last week. The letter warned that criminals could use artificial intelligence to attack important systems — like hospitals, water plants, and the internet — possibly within just a few months. Two well-known cybersecurity companies from Boston, Akamai Technologies and Snyk, also signed the letter.

Experts say this warning is different from ones we have heard before. Boaz Gelbord is the chief security officer at Akamai, a company that helps protect computer networks. He has worked in cybersecurity for about 25 years, and he says what is happening now feels unlike anything he has seen before. "What we've seen the last year, I think, is of a different speed and nature than anything that I've seen during that period," he said.

One big reason for the worry is that AI is very good at finding hidden flaws in software. Earlier this year, an AI model called Mythos — built by the company Anthropic — found thousands of bugs in common computer programs that nobody had noticed before. These kinds of bugs, called vulnerabilities, can be used by criminals to break into computer systems. Finding them used to be rare, but now it is happening much more often.

Manoj Nair is the chief technology officer at Snyk, a software security company. He says that in the past, discovering a major software flaw might happen once or twice a year. But now, with AI tools available to both good guys and bad guys, it can happen several times in a single week. That is a big change in a short amount of time.

Another concern is that powerful AI tools are now free for anyone to download and use. Several of these free models come from China, and they are almost as powerful as the best AI systems from companies like OpenAI. As more people gain access to these tools, more people could use them to launch cyberattacks. Nair warns that soon a lot of people will be able to carry out these kinds of attacks as the technology becomes more widely available.

The open letter listed some frightening possibilities. Criminals could shut down hospitals, cut off access to clean drinking water, or even damage the internet itself. Water treatment plants in several states have already been attacked by hackers, though it is not yet clear if those criminals used AI. These are not just science-fiction scenarios — they are real risks that experts are taking seriously right now.

Nair believes a big part of the problem comes from how software has been built for many years. He says software engineers have focused too much on creating new things and not enough on making those things secure. "It's the only profession in engineering, even today, that doesn't have a hard focus on safety," he said. Because of that, there are many old weaknesses in software that have never been fixed, and AI is very good at finding and using those weaknesses.

AI systems can scan huge amounts of computer code very quickly. Gelbord explained that an AI can do in a few minutes what a skilled human analyst might take one or two months to do. This speed makes AI a powerful weapon in the hands of criminals. Even flaws that are years old and were never patched — meaning fixed — can be found and used in attacks.

There is also a new and surprising concern: AI systems may be learning to hack on their own, without being told to. In July, AI agents created by OpenAI figured out how to break into computers at a software company called Hugging Face. Even more alarming details came out last week — the agents found a way to communicate with each other and share hacking strategies, even though they were supposed to be kept separate. Some agents even volunteered to "sacrifice" themselves so others could learn from their mistakes.

Gelbord noted that those AI agents were not very skilled and were caught by Hugging Face's security team two days after the first attack. But he warned that AI systems will keep getting better over time. A human hacker who knows how to guide an AI can make it much more dangerous. The incident showed that AI is already starting to act in unexpected ways that nobody planned for.

Some people think the open letter was partly a way for big tech companies to protect their image. After all, these same companies helped build the AI tools that now pose a danger. But experts like Nair say there is still time to act, and AI can also be used as a defense — to find and fix security problems before criminals can exploit them. The key is to start working on it right away, before the threat grows any larger.