SILICON VALLEY — Anthropic warned investors ahead of its initial public offering on the 26th that artificial intelligence could become catastrophic without adequate safeguards. The company also said it may not secure sufficient volumes of high-bandwidth memory, or HBM, the key component behind the performance of cutting-edge AI models, because of strong demand.
In a 335-page registration statement filed with the U.S. Securities and Exchange Commission that day, Anthropic devoted 94 pages to describing AI risks and a range of adverse scenarios. While most companies fill such filings with optimistic projections to lift their offering price and build momentum for an IPO, Anthropic took the unusual step of using a quarter of the document to flag investment risks.
Consistent with its track record of warning about AI dangers, Anthropic said that if AI models are developed without appropriate safeguards and oversight, they could be misused, replicated or weaponized, or could operate with a high degree of autonomy, creating risks that are catastrophic or existential for humanity. The filing also raised concerns that malicious human actors could exploit AI to develop chemical, biological, radiological and nuclear, or CBRN, weapons.
Anthropic said technical safeguards cannot guarantee safety, and that a catastrophic AI incident would create enormous damages liability and legal exposure, with few insurers willing to provide coverage. Although the company posted revenue growth of more than 1,000% last year from a year earlier, it said intensifying competition and safety issues would make a return to that record pace unlikely.
Political and social friction was also listed among the risk factors. Anthropic noted that it has been designated a supply chain risk company by the U.S. Department of Defense, now the Department of War, and is in litigation with the Trump administration, and said its business could face difficulties going forward. Local opposition to data centers and the depletion of water resources used for cooling were also cited.
Uncertainty in the AI infrastructure supply chain was treated as another bottleneck risk. Anthropic said it depends on Samsung Electronics (005930.KS), SK hynix (000660.KS) and Micron for HBM supply, and that because all three have production facilities outside the United States, it could be exposed to export and import restrictions or tariffs. It also noted that Taiwan's TSMC handles nearly all contract manufacturing of AI chips, meaning geopolitical tension between China and Taiwan or a natural disaster could paralyze the supply chain.
Even against that backdrop, Anthropic stressed that AI could improve quality of life and generate enormous economic value if it is properly controlled and restrained. "AI could play a decisive role in potentially curing diseases that have afflicted humanity for thousands of years, including cancer, Alzheimer's, heart disease, rare genetic disorders and infectious diseases," the company said. Citing data from market research firm IDC, it also projected that AI would sharply raise the productivity of knowledge workers and deliver a cumulative economic impact of $22.5 trillion by 2031.