Anthropic, the AI safety company behind the Claude family of models, is facing a wave of backlash after a policy shift caught its business partners and industry allies off guard, according to reporting by The Information.
The fallout has reached unexpected corners of the AI world. An OpenAI employee who had joined colleagues in filing an amicus brief supporting Anthropic in a legal dispute against the Pentagon announced he is withdrawing his signature, according to reporting by MSN. The move signals that the reversal has strained goodwill even among rivals who had chosen to stand with Anthropic.
Meanwhile, Tom's Hardware reports that Anthropic has issued a warning that AI self-improvement could eventually lead to humans losing control of frontier AI models — a striking message from a company that only weeks earlier declined to release an internal model called Mythos, describing it as too powerful to make public.
Tom's Hardware notes there is a subtext to Anthropic's warning: the company appears to be arguing that accelerating AI development requires significantly more compute before the risk of losing control becomes real — framing safety concerns as a reason to invest more, not slow down.
The back-to-back moves have left observers parsing whether Anthropic is genuinely recalibrating its safety posture or sending a strategic signal about resources and regulatory positioning.
This matters because Anthropic has long positioned itself as the measured, safety-first voice in a field dominated by aggressive builders — and any perception that it is shifting that stance, or doing so without warning partners, could reshape alliances and trust across the entire AI industry at a critical regulatory moment.