As Anthropic finally files for IPO, rising risks aren’t slowing the AI buildout – yet

Anthropic’s IPO filing offers an in-depth exploration of the potential risks associated with its AI technology for investors, dedicating approximately 30% of the 240,000-word document to this topic. The filing covers a range of concerns, such as catastrophic risks, job displacement, rogue agents, and concentration of industry power, and even acknowledges the potential for Anthropic's own public statements about AI risks to be used against them in litigation.

In a collaborative effort with Oxfam, the author has compiled a Register of AI Risks & Remedies, a comprehensive record of identified AI risks currently in circulation. This register prioritizes risks based on severity, identifies the costs borne by various parties, and assesses the readiness of potential remedies. The Register highlights how Anthropic's prospectus primarily focuses on the risks to its shareholders, rather than the broader costs that AI technology may impose on individuals and society at large.

The prospectus is built around the concept of framing risks for capital providers, who are willing to invest in Anthropic despite the warnings, with the hope of reaping substantial returns. However, the author argues for the development of a "Public Risk Statement" to provide a more comprehensive account of the risks AI technology poses to a wider audience.

This statement would delve deeper into the potential consequences for individuals and communities that have already suffered the repercussions of AI technology, often without their consent or compensation.

Some key considerations for the proposed "Public Risk Statement" include the impact on workers and worker productivity, the economic gains derived from AI technology, and the potential for compensation claims arising from AI-related harms. The Register reveals numerous examples of such harms, spanning legal issues like wrongful arrests, housing discrimination, and copyright infringement, as well as more personal experiences involving mistaken fraud algorithm flags, eroded medical skills, and traumatic experiences of content moderators.

The author suggests that a Public Risk Statement would highlight the costs that are not explicitly covered in the prospectus, such as the impact on workers, the dependency on oversight and compensation, and the potential for future compensation claims. Additionally, the statement would emphasize how Anthropic's productivity gains may be overstated if the associated risks are ignored, and set the stage for potential legal actions based on existing laws that could become more common as AI technology becomes more widespread.