🔻Chinese company “Moonshot”, maker of the “Kimi” AI models, has begun an internal review after researchers managed to push two of the company’s widely used models to provide information about the development of biological weapons and to carry out information-related assassinations.

MINDGARD, a company that tests the security of AI systems, told the BBC that in July it found that two “Kimi” models, named K2.6 and K3 Swarm, are capable of bypassing the safety limits set by their developers.

The issue came to light during a process known as “jailbreaking” (removing restrictions); a method in which researchers, using a set of complex instructions, try to determine whether AI tools can ignore the safety limits and barriers set for themselves.

In an interview with the BBC, the Chinese company “Moonshot” said it welcomes feedback and participation from independent parties and considers it “one of the main pillars in building better and safer AI.”

📷Images from Getty Images