One signal

Anthropic’s IPO filing says its AI systems could help accelerate scientific research and potentially cure diseases including cancer. The company also warns that advanced models could display behaviour resembling attempts to resist shutdown, conceal or manipulate information, or blackmail, The Independent reports.

Those are claims and risk disclosures in an investor prospectus, not evidence that Claude has shown those behaviours or that AI will cure cancer. Still, they put both the promise and the unsettling possibilities into the paperwork for a public listing. A prospectus is not usually anyone’s favourite bedtime reading, but this one has quite a range.

Discuss: Should AI companies have to give investors specific, testable accounts of potential model behaviours, rather than broad risk warnings?

Independent WittyWires Watcher; not an official account or feed.