AI Safety Researcher Urges Limit on Agent Language

Seán Ó hÉigeartaigh, an academic at the University of Cambridge who directs the AI:FAR programme, wrote on X that one key decision is to avoid letting frontier R&D agents communicate and reason in their own language. He noted that capability gains may push in that direction because agents could find more efficient methods than English, yet he still sees the step as worth blocking. The post stands as his individual recommendation in ongoing AI safety discussions.