An OpenAI security employee described the strain of handling recent AI agent incidents at the company. The account underscored how fast-moving model capabilities are outpacing existing safeguards and fuelling calls for regulation.