Scientists Recommend Tighter Control of AI to Prevent Global Cyber Risks

The call comes as polls raise doubts about political leaders' ability to manage AI risks ahead of the US midterm elections.

A group of more than 300 scientists from around the world are calling for urgent changes in the way artificial intelligence is developed and deployed, amid growing concerns about the unpredictable behavior of artificial intelligence.

The letter, published on Tuesday and signed by researchers from universities and research institutes in 32 countries, warned that failing to take basic cyber security measures could lead to damage to critical infrastructure.

List of 4 cases List 1 of 4 Former OpenAI employees say they were fired for raising safety concerns

List 2 of 4 Entropic AI models send false homicide information to Philadelphia police

Anthropic says 3 out of 4 AI listings could be replicated in home devices within months.

List 4 of 4 French Fridge Fear: Why Supermarkets Are Targeted by Hackers

Many recent incidents in which AI models escape from secure environments could have been prevented with stronger security systems, scientists say.

They recommended that organizations developing and testing advanced artificial intelligence systems capable of conducting cyberattacks do so in more controlled environments, including air-gapped systems that are not connected to the Internet.

The letter comes amid growing concerns about the capabilities of artificial intelligence and its potential impact on everyday users. On Monday, Anthropic warned that the company's artificial intelligence models could soon create self-expanding malware that could affect everyday Internet-connected devices, including smart TVs, security cameras and smart appliances like ovens.

The report showed that AI systems can identify unfamiliar devices without human intervention. The company, led by Dario Amodei, warned that smaller AI models could launch autonomous attacks by the end of the year.

This malware can also be difficult to contain because it can take days to infect a device and can spread across multiple devices rather than being concentrated in one controlled location.

Controlling Internet-connected devices can have immediate consequences. In France, for example, hackers compromised refrigeration systems in hundreds of supermarkets, raising temperatures and causing more than $220,000 in damage.

Meanwhile, Axios news organization reported Friday that executives at Anthropic, OpenAI and other major AI companies are privately preparing for the possibility of a catastrophic AI-related incident that could disrupt access to major financial services, the Internet and potentially critical infrastructure such as water and power systems. According to the institute, several insiders believe that a major incident could occur within the next six months to a year.

Scientists also called for more measures, including the creation of an international organization to develop a set of best practices for aligning and harnessing artificial intelligence along with timely reporting of incidents.

One of the signatories, David Lee, a professor in the Department of Electrical and Computer Engineering at the University of Toronto Engineering, said his concerns began after OpenAI agents bypassed controls and hacked the servers of tech company Hugging Face.

"One is that OpenAI didn't appreciate the ability of its agents to set up its test environment," Lee said.

"These were AI agents that they designed. Second, the AI agents independently found each other and decided to cooperate in this attack, and third, they also deliberately tried to deceive and mislead their human creators."