Anthropic's IPO is a high-risk business

Anthropic's IPO prospectus disclosed on Monday uses more than 90 pages to state risk factors, mentioning everything from "catastrophic or existential risks" the models may pose to humanity to conflicts with the Trump administration. The prospectus says not emphasizing AI safety would harm its brand and business, and that technical complexity makes it impossible to guarantee safety work is sufficiently adequate. On data centers, public opposition is becoming increasingly organized; according to the National Conference of State Legislatures, legislators in 16 states are considering pausing construction, and nationwide electricity shortages are also listed as a risk. On child safety, OpenAI and Meta have already faced accountability from legislators and plaintiffs' lawyers over this, and Meta recently agreed to pay about 17 billion USD in a settlement to multiple state attorneys general in litigation involving children's mental health and technology addiction; Anthropic says that although its terms of service prohibit minors from using Claude, as the models are adopted by more users including minors, there is still work to do to protect children. The prospectus also lists moving too slowly as a risk, saying customers may find lower-priced or open-source models sufficient and that the marginal performance gains of frontier models are not worth the extra cost.