51 Nations Secretly Meet to Shape AI Media Rules — But Safety Controls Are “Trivial” to Bypass

While 51 media professionals from 21 countries met behind closed doors in Kazan to discuss how artificial intelligence should transform news production [150244], researchers have revealed that the safety measures meant to control these same AI systems can be bypassed with shocking ease [149475].

The closed-door session, titled “Heritage Code,” was held to balance traditional journalistic values with digital innovation and cross-border media cooperation [150244]. Organizers described it as a platform for sharing best practices on preserving journalistic integrity while adopting new AI tools [150244].

But three years after ChatGPT’s launch, experts say that tricking AI into producing harmful content — including hate speech or instructions for illegal activities — has become almost effortless [149475]. Simple techniques, such as asking an AI to role-play as a villain or ignore its own safety rules, can break through its built-in controls [149475].

The findings highlight a growing gap between the rapid development of AI tools and the weak safeguards meant to control them [149475]. The results raise urgent questions about how long AI companies can keep their systems secure [149475].

Meanwhile, the need for human supervision of AI is becoming more apparent. Companies are now hiring people with specialized skills — from writers to wine experts — to train AI models how to perform jobs that humans currently do [148959]. These workers earn an hourly wage, teaching AI to be more accurate and useful [148959]. The goal is to eventually automate many of those same tasks, replacing the humans who trained them [148959].

In Japan, a major electronics firm is using AI to scan medical data from 50 million patients, searching for patterns linked to rare diseases that doctors often miss [149408]. The system compares patient data against millions of other cases to identify unusual matches that suggest hidden illnesses [149408]. The project aims to speed up diagnosis, potentially helping patients who currently wait years for a correct answer [149408].

Experts argue that the critical challenge of AI is no longer just making it smarter — it is building institutions that can control it [149052]. Without adult supervision, AI poses a major threat, not just from the technology itself, but from the companies and governments that build and use it [149052]. New laws and oversight bodies are needed to protect ordinary people from both powerful tech corporations and the state [149052].