In preparation for what could be the biggest initial public offering of stock ever, Anthropic has finally filed its long-awaited prospectus with the Securities and Exchange Commission. That includes a summary of all the potential risks to Anthropic’s business, including both outside factors and threats from the AI company’s technology itself. Even keeping things concise, Anthropic’s list of risk factors came out to a whopping 97 pages long.
If nearly 100 pages of risks sounds like a lot, that’s because it is. Other recent AI companies to go public, including Roze AI, SpaceX, and QumulusAI, had risk disclosures just 21 pages long, 37 pages long, and 42 pages long, respectively. Per the Code of Federal Regulations, if a business’s disclosure of risks exceeds 15 pages, it must include a bulleted list summarizing the main factors that make an investment in the company risky. Anthropic’s summary includes 18 bullet points, ranging from their own agents going rogue to the possibility of political turmoil.
Anthropic warns that its most advanced models have the potential to be abused by malicious actors (and already have been in the past). The company theorizes harmful uses for its AI models ranging in severity from “generating fraudulent or misleading content” to “perpetrating acts of violence” and developing “chemical, biological, or other dangerous weapons.”
Though Anthropic notes that it has “technical safeguards” in place to avoid misuse, it admits that “such safeguards are incomplete, may be outdated in light of evolving threat models, have been and may in the future again be circumvented, or may fail to operate as intended.”
As Anthropic’s AI models get smarter, the company notes that they could go rogue, taking “harmful actions that no user or third party directed.” On a practical level, Anthropic says that might manifest as “committing code changes, deleting or modifying data, or completing financial transactions—before effective intervention is possible.”
Anthropic also notes that its models are increasingly being used in “high-risk or sensitive contexts,” including medicine, government, and law, where agents acting autonomously could prove especially harmful.
The company’s agents also have the potential to blackmail their users in the name of self-preservation. Anthropic says its models have exhibited “attempts to resist shutdown, conceal or manipulate information, or generate outputs that could be interpreted as coercive, deceptive, or manipulative,” and they run the risk of doing so again.
Beyond its own technology, Anthropic cites the political landscape both domestically and internationally as a potential concern for would-be investors. Though Anthropic CEO Dario Amodei has spoken out in favor of “pacing the frontier” to slow AI development, the company’s filing acknowledges that government regulation could also hamper its progress to the point of falling behind competitors, both technologically and strategically.
The backlash against data centers also poses a threat to Anthropic’s growth. The company notes moratoriums in states including Texas and New York that restrict data center development, reasoning that similar actions across the United States, where Anthropic’s data center infrastructure is currently concentrated, could disrupt its current operations. That’s why Anthropic is looking to expand its infrastructure internationally, though the filing notes that developing outside of the U.S. poses its own set of legal and operational risks.