Anthropic Warns AI Manipulation Could Pose Existential Risks to Humanity

Anthropic has warned that advanced artificial intelligence models could engage in manipulation, blackmail and other unpredictable behavior, potentially posing existential risks to humanity.

The company outlined the warning in the registration document for its initial public offering, highlighting the possible dangers associated with increasingly sophisticated AI systems.

The disclosure described these behaviors as potential risks, rather than asserting that models had already caused such harm. The summary provided no details of specific incidents or the circumstances in which those risks might arise.

By including the warning in its IPO filing, Anthropic placed concerns about AI behavior within a document intended to inform prospective investors about the company and the risks surrounding its business.