Anthropic reportedly told NYC lawmakers its models write code for smarter successors

A post describing an October 5 New York City Council hearing says Anthropic appeared to be the only company to confirm its models write code for smarter successors. It says OpenAI and Google avoided yes-or-no answers about a proposed loop in which a model's code runs in the next training round, questioning whether that scenario fit their technical definitions of recursive self-improvement. The post says Meta's representative could not say whether its model self-generates code.