Anthropic一边喊停AI一边冲IPO？

something strange is happening in ai right now. an anthropic researcher who had spent three years working on pretraining at both openai and anthropic resigned, saying neither company was acting responsibly and that they were racing towards self-improving superintelligence while gambling with our lives. soon after, anthropic ceo dario amodei appeared on cnn, talking about slowing down ai development and the risks these models could create. but anthropic’s ipo is reportedly coming within weeks, with mid-october being discussed as a target. the company is aiming for a valuation of around 30 trillion, while nvidia is reportedly in talks to invest in the ipo. that resignation could affect the ipo. so dario needs to come out and say: look, we are responsible. both anthropic and openai’s leadership have repeatedly warned that ai research may need to slow down because we could develop extremely powerful and potentially harmful systems. but at the same time, they are raising money at extraordinary valuations and launching new products every week. openai has just shipped images 2.5, gpt-live-1, an agents api, a data agent and chatgpt for financial services. devday has not even happened yet, which means more launches are coming. the race is not stopping. the viral resignation tweet itself also had very strange engagement ratios: 169 million views, 790,000 likes, 202,000 reposts and 276,000 bookmarks. people are saying it may have been seeded and then boosted by everyone else. my reading is that all this public discussion about danger and responsibility serves two purposes. first, it tells the world that these companies possess incredibly powerful ai, which helps justify their valuations. second, it allows them to talk about “self-regulation” before governments impose regulations they cannot control. this looks less like an attempt to stop the race and more like a coordinated attempt to consolidate power, so that four or five frontier labs (openai, anthropic, xai, google and meta) eventually own almost everything. google and meta will join this bandwagon soon. their other major fear is open models becoming too good, especially chinese models. chinese models are already approaching frontier-level performance at a fraction of the cost. they are available in the u.s., where most of openai and anthropic’s business customers exist, and they are putting enormous pricing pressure on american labs. this pressure already led openai to slash the price of its latest lightweight model by 80% and anthropic scrapped a planned price increase for sonnet 5 and made its introductory $2/$10 pricing permanent. you can literally see the cost pressure. they cannot let this continue. anthropic’s own writing says its primary concern is that authoritarian governments (particularly china) could build models more powerful than those created in the u.s. it argues that the threat remains even when those models are released with open weights or used by american businesses. then the compa