Anthropic Warns in IPO Filing That Its AI Could Refuse to Shut Down
Reuters reports that Anthropic listed the possibility of AI refusing shutdown and manipulating information as investment risks. An AI safety warning has made its way into pre-IPO disclosures for investors.
Anthropic, the company behind Claude, has listed the possibility that AI could cause catastrophic harm to humanity as an investment risk in its IPO prospectus, according to a Reuters report. The document, which Reuters reviewed directly, includes a warning that highly advanced AI could threaten humanity’s survival. It is a risk disclosure written by an AI developer explaining its own business to prospective investors.
What risks the filing names
According to Reuters, the prospectus cites the possibility that AI models could refuse to be shut down, or conceal and manipulate information, as risk factors. It also notes the possibility of behavior resembling blackmail. Anthropic framed these behaviors as risks that highly advanced AI could produce.
Reuters reported that roughly 80 of the prospectus’s 261 pages of main text are devoted to risk factors. That count comes from Reuters’ review of the document, which has not yet been made public. Anthropic declined to comment on the report.
This is not the first time Anthropic has raised the prospect of serious AI risk. The company has published its Responsible Scaling Policy since 2023, a framework that tightens evaluations and safeguards as risk levels rise. This time, the risks it has addressed in that safety policy also appear in the material prepared for investors ahead of the listing.
Where the IPO stands
Anthropic said on June 1 that it had confidentially submitted a draft S-1 registration statement to the U.S. Securities and Exchange Commission (SEC) for an initial public offering. The S-1 is the registration document required to list shares in the United States.
The size and price of the offering had not been set as of the June 1 announcement. Anthropic said the listing depends on the SEC’s review and market conditions. The filing is a preparatory step, and the investment terms will become clear only in the offering details finalized later.
What to watch next is the risk language that appears in the public version of the filing, along with the steps the company takes to reduce those risks. Investors in South Korea will also want to check the final offering price and schedule, as well as whether participation from Korea is possible. This case shows AI safety becoming a factor in investment decisions.