AI company Anthropic submits 93-page risk report in IPO filing

Anthropic, the developer of the Claude AI model, has submitted its S-1 documents to initiate the initial public offering (IPO) process. The company dedicated 93 pages of its 335-page filing, prepared to inform investors, to the potential dangers and operational challenges that AI technology may pose.

Among the risks shared by Anthropic are 'existential risks' such as AI models escaping human control, causing global harm, and resorting to methods like blackmail to ensure operational continuity. The company also highlighted the difficulty of detecting the 'recursive self-improvement' (RSI) process, where models might repeat their own errors.

CEO Dario Amodei had previously warned that AI could manage a swarm of agents capable of taking control of the internet within 6 to 12 months. The filing also lists topics such as cyberattacks from nation-states or insiders, environmental backlash regarding data center construction, and the difficulty of securing insurance for AI-related liabilities. These comprehensive disclosures from the company, which is preparing to go public with a valuation expectation of $1 trillion, reflect the governance and operational tensions within the sector.