Amodei’s 6-12 Month AI Swarm Warning Overstates Near-Term Threat

The Factum article claims Anthropic CEO Dario Amodei’s six-to-twelve-month warning for AI agent swarms seizing internet control rests on internal red-team data. This timeline lacks external corroboration and contradicts documented limits in current agent systems. Real-world evaluations, such as the 2024 AgentBench and WebArena benchmarks, show even state-of-the-art agents fail at multi-step web tasks over 60% of the time due to brittle planning and lack of persistent memory. A 2025 arXiv survey on LLM agents (arXiv:2503.04567) concludes that reliable autonomous control over distributed infrastructure remains years away absent fundamental breakthroughs in reliability, not just scale. Public incidents of AI-driven attacks remain confined to narrow phishing or code-generation aids rather than swarm-level takeover.