More than 400 researchers from leading universities and AI laboratories signed a letter this week calling for mandatory safety evaluations of all AI systems above a certain capability threshold. The letter warns that current development practices prioritize capability gains over safety analysis, creating systems whose behavior cannot be reliably predicted or controlled in all circumstances.
The call has divided the AI research community. Some prominent figures argue the risks are overstated and that safety restrictions will cede ground to less cautious developers in other countries. Others contend that the industry's track record of self-regulation is insufficient given the potential stakes. Congressional hearings scheduled for next month will feature testimony from signatories on both sides of the debate.
