Before the penetration of “self-replicating” AI in gadgets remained a few months

Once that happens, the models will be able to connect to these devices on their own without using data centers. Such self-replication would lead to a serious loss of control over the unmanageable model of artificial intelligence.

“A replicator like this would be difficult to disable,” the anthropologists wrote in a report published Monday, “because there is no centralized point to shut down.”.

The company said these AI worms, a kind of standalone malware, could have such capabilities “in a few months.” Once out-of-control AI systems are able to hack smart devices, they can spread across networks without detection. In the worst-case scenario, various AI worms could “combine” to attack computer hardware, according to the company.

“We should strive to strengthen our systems to make it difficult for such replicators to spread, but also prepare to deal with a world in which they will remain forever,” the researchers wrote.

The new Anthropic report comes as AI security concerns have reached their climax, The Guardian comments. Over the past few months, there has been a steady stream of reports from AI companies including Anthropic and OpenAI that their agents are capable of acting independently and hacking other companies and government agencies.

These revelations have led researchers, lawmakers and executives, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, to call for a slowdown in the industry's growth rate and increased precautions. Despite this, both companies continue to actively develop and produce new models.

The Anthropic team, which stress-tests artificial intelligence systems, produced a report published Monday. The idea behind the team is to explore the possibilities of their artificial intelligence models. In a scenario with self-replicating AI worms, the researchers asked the agent to hack devices connected to the Internet.

They found that AI self-replication is likely to work only with smaller models and that homes and small businesses are likely to be the first targets for such attacks. All the devices that Anthropic considers vulnerable are equipped with small built-in artificial intelligence chips, including smart ovens, smart TVs, automatic lawn mowers, CCTV cameras and video conferencing devices.

The researchers concluded that it was unclear what form AI replication might take. They said it could be a slow model, such as a “digital mold,” a model that tricks people into getting credit for computing, or a “powerful cyber model” that quickly hacks into large objects like data centers.

However, anthropologists say they expect a future that is “rife with agents working on all types of equipment, including those not under human control.”.