Anthropic should publish a separate “Public Risk Statement” setting out who may be harmed by its products and how affected people can seek redress, according to an analysis of the artificial intelligence company’s planned public offering.
The proposal argues that the company’s prospectus focuses largely on risks to investors and the business itself, rather than the wider social and financial costs that may be borne by workers, consumers and communities.
Anthropic’s 240,000-word filing devotes about 69,800 words to risks, covering issues including job displacement, rogue AI agents, catastrophic harm and the concentration of power in the industry.
It also warns that the company’s own public statements about AI risks “may be used against us in litigation”. However, the analysis says the document does not fully account for the potential costs of Anthropic’s products to people outside the company.
Anthropic is a Delaware Public Benefit Corporation, whose charter requires it to balance shareholders’ financial interests with the benefit of the broader public. Delaware law also requires periodic reporting on how the company has served people materially affected by its conduct.
The company’s existing disclosures describe its efforts in that area, the analysis says, but do not sufficiently detail their outcomes. A Public Risk Statement would instead identify the uses of its products, those who might be harmed, responsibility for those harms and the practical routes available to people seeking compensation.
Workers and hidden productivity costs
A Register of AI Risks & Remedies compiled in collaboration with Oxfam contains 103 documented risks, ranked according to their severity, who bore the cost and whether remedies were available.
The register identifies a recurring risk of over-reliance on AI and the erosion of workers’ skills when automated systems gain authority and the people supervising them lose the ability to challenge their decisions.
Anthropic’s prospectus refers to over-reliance only once and does not mention deskilling, according to the analysis. It acknowledges that human oversight can fail when workers are under pressure to improve productivity.
A separate statement would need to show who checks AI outputs, whether they have enough time and information to do so, and whether they are empowered to overrule a system. Claimed productivity gains could be overstated if the cost of checking work is excluded.
The analysis also warns that financial returns could be weakened if productivity gains depend on poor oversight or if compensation later has to be paid to people wrongly affected. Anthropic says its monitoring “does not cover all” AI deployments, but the filing does not set out the full range of possible exposure.
Potential liability for AI companies
Previous disputes illustrate how costs can arise from the use of automated systems. Detroit paid Robert Williams 300,000 dollars after a flawed facial-recognition match led to his wrongful arrest, while tenant-screening company SafeRent agreed a 2.275 million dollar settlement over discriminatory screening.
Anthropic itself agreed a 1.5 billion dollar settlement to compensate authors and publishers over pirated books. The examples span civil rights, housing and copyright, and the analysis argues that wider adoption could expose more people to mistakes and increase demands for compensation.
It says that around four in five of the 103 harms recorded in the register involved an existing law that could apply, suggesting the immediate gap may be less about the absence of legislation than the practical difficulty of securing redress.
That gap could narrow as legal and regulatory systems develop. Legislators in 30 US states enacted 115 AI-related laws in 2026, according to the Transparency Coalition’s report, while eight state attorneys-general are suing OpenAI over alleged harm to children.
The Federal Trade Commission is investigating safety lapses at frontier AI laboratories, the European Union is beginning enforcement of its AI Act, and the Financial Stability Board is urging banks to improve AI governance.
The proposed Public Risk Statement is not required by law, but its supporters argue that it would give investors a clearer picture of how much of Anthropic’s business depends on costs carried by others. They also warn that shareholders could eventually bear those costs if claims for compensation gather pace.
