AI leaders believe the industry cannot self-regulate after initial warnings
Read more
CGTN
cgtn.com

AI leaders believe the industry cannot self-regulate after initial warnings

Two leading figures in artificial intelligence have stated that the systems they created are beginning to spiral out of control, emphasizing that no company should operate outside the legal framework.

Demis Hassabis, co-founder of Google DeepMind, called recent hacking attacks carried out by AI agents 'early warning shots.' He warned that if the systems do not become more secure, the world could face serious incidents within one or two years.

At the AI summit in Dublin, Ireland, on Wednesday, Hassabis, who is also the chief scientist at Google's parent company, Alphabet, noted: 'I think these (hacking incidents) are early warning shots for us, showing that the systems are not yet powerful enough, so the incidents are not too serious.' However, he added that the situation could become serious next year and in the following couple of years if the industry, individual companies, and governments do not take measures to ensure the security of these systems.

Microsoft President Brad Smith reported at the conference that the uncontrolled behavior of AI has made 2026 'anxiety-inducing,' and that new laws are urgently needed for safety to keep pace with innovation.

He stated: 'No company or technology should be above the law. But we need laws, we need regulation. It must be reasonable, balanced, and take innovation into account.' Smith also expressed the view that the resistance to regulation shown by some industry representatives is a problem that needs to be overcome.

Smith urged the industry and governments to 'act decisively' to create 'real and effective safety standards.' He stressed that AI requires the same real and effective safety standards as everything else in life, as well as public trust in them.

The Microsoft boss presented proposals to his company, including a requirement that AI agents have 'safety brakes' that users can activate in case of an incident. He specified that this capability is especially important for any AI operating on critical infrastructure or any other automated system, allowing it to be shut down or slowed down.

Popular