The call coming from inside the house

The accusation that Anthropic dresses competitive advantage as safety is old. What changed is who is making it. Nathan Lambert runs post-training at the Allen Institute for AI and writes one of the field's most-read technical newsletters; he is the mainstream of open-model research, not an e/acc partisan. In late June he posted that Anthropic's political push against distillation is regulatory capture — and that most of the company's employees are 'blind to it under their veil of safety.'

The legible proof is a number. Anthropic spent roughly $1.56 million on federal lobbying in the first quarter of 2026 — its biggest quarter ever, up from about $360,000 a year earlier, and more than OpenAI. In the same stretch it asked senators for sanctions on Chinese distillers, tighter chip export controls, and an antitrust carve-out so frontier labs can coordinate. A firm that stands to gain from the remedy is supplying the alarm; that is the recognisable shape of capture, whatever the intent behind it.

Lambert is careful about the intent. He grants that leadership may not have set out with regulation as the goal — distillation, he argues separately, is an industry-standard technique, and an over-broad ban would gut the US open-weight ecosystem long before it dents China's. The sharper claim is the one about the staff: that a safety mission, sincerely held, is exactly what makes a competitive moat invisible to the people building it. The irony is that four months earlier Anthropic quietly deleted the one unconditional pause clause from its safety policy — safety strong enough to lobby with, not strong enough to bind.