Amodei puts a 6-12 month clock on an agent swarm that could run a persistent internet-wide botnet, and opens Anthropic to outside evaluators

VentureBeat·high signalXBlueskyLinkedInCopy link

In a personal essay posted 2026-09-12, Dario Amodei wrote that within 6 to 12 months an AI swarm 'could be capable of taking over the entire internet with a persistent botnet,' potentially causing hundreds of billions of dollars in damage. Alongside the warning Anthropic unilaterally committed to the first step of a three-part slowdown plan: permanent employee-level access for third-party AI safety evaluators, who keep publication rights without company editorial control apart from narrow security and legal redactions. The timing sits directly against the company's own threat report showing criminal agent swarms already operating, which makes the claim harder to read as pure positioning.