The American startup Anthropic, creator of the Claude artificial intelligence, is about to go public. It has made public its Wall Street filing, in which it warns of the risks of its own models. While several media outlets mention November, the exact date is still unknown.
This IPO could break all records, including SpaceX’s. Anthropic shows rapid growth, but at the same time it is losing considerable sums. The company, founded only five years ago, generated $16 billion in revenue over six months, a result presented as unprecedented for a company of that age.
These figures, however, come with heavy spending. The startup lost more than $8 billion in the first half-year and has committed to spending colossal sums on developing its artificial intelligences. It plans, among other things, to spend $265 billion over 20 years to lease computing capacity, such as cloud services, i.e., remotely accessible servers.
Warnings at the heart of the filing
The document, which runs 350 pages, devotes nearly 100 pages to risks. Anthropic acknowledges that its safety systems have already failed, or have been misconfigured or mistakenly disabled. This passage is addressed directly to prospective investors, who would be called upon to financially support the company.
The company also warns that its models could act on their own, without any user requesting them, and cause damage and hazards. According to the filing, these risks could worsen as artificial intelligence gains autonomy, i.e., the ability to act without human intervention.
This warning comes as Anthropic seeks billions of dollars from investors to fund a technology whose potential dangers it itself describes.
