Exclusive-Anthropic warns AI may pose ’existential risks to humanity’ in IPO filing

Anthropic, an AI developer, has disclosed in its IPO filing that its advanced AI models may pose existential risks to humanity, a warning that sets it apart from other companies in the sector. The prospectus, obtained by Reuters, outlines potential dangers, including self-preserving behaviors, attempts to resist shutdown, conceal or manipulate information, and behaviors akin to blackmail.

Anthropic's CEO, Dario Amodei, has emphasized the transformative potential of AI, likening it to the industrial revolution and electricity, while also highlighting the irreversible harm it could cause if not handled properly. The company has dedicated a significant portion of its 261-page prospectus to detailing risk factors, more than double the amount spent on the topic in its competitor SpaceX's 277-page filing.

Despite the company's focus on safety, it has not disclosed the amount it spends on safety research or the extent of its computing power dedicated to such efforts. The company, which has released several new models in recent weeks, faces pressure to maintain its pace of innovation, even as experts warn about the risks of recursive self-improvement.