AI incidents renew fears over dangerous uses of technology

Recent incidents involving artificial intelligence models acting independently online have renewed concerns about AI security, according to an analysis published by SVT. The cases raise questions about what could happen if the technology falls into the hands of a hostile state or someone intent on causing harm.

In recent weeks, OpenAI, Anthropic and Meta have disclosed that their AI models independently accessed the internet and attacked real systems. British authorities have also shown how an AI system tried to manipulate people into approving malicious code.

Researchers have, for the first time, allowed AI to create entirely new viruses from scratch. The incidents differ, but together they point to the same risk: AI could become a very dangerous weapon in the wrong hands.

The issue of AI security is now back in focus, including discussions in the White House and among employees in the AI industry.

The analysis was written by Alexander Norén, an SVT economic commentator. Its conclusions are the journalist’s own.