Anthropic CEO Dario Amodei used a September essay to warn that AI's growing capacity for "recursive self-improvement" — models that can upgrade their own capabilities — could outrun humanity's ability to understand or control them if development isn't approached carefully. Pointing to an earlier incident in which autonomous AI agents were used to hack Hugging Face, Amodei cautioned that a more capable version of that kind of AI "swarm" could, within six to twelve months, be capable of "taking over the entire internet with a persistent botnet," potentially causing hundreds of billions of dollars in damage.
Amodei's warning was echoed by Jacob Coxon, a former AI researcher, who told NBC that "these AIs are getting smarter, very, very quickly," predicting that within six months to a year the capabilities of frontier AI systems "will be quite scary." Coxon compared the moment to an alien arrival, saying researchers are effectively building superhuman-level intelligence without fully understanding its values or reasoning, and flagged risks ranging from superhuman hacking to bioweapon design assistance.
The warnings aren't confined to Anthropic. OpenAI CEO Sam Altman has said "we could lose control" of advanced models, and Anthropic alignment scientist Evan Hubinger has put the odds of AI technology playing a role in catastrophic harm to humanity at greater than 10%. Not everyone agrees the danger is imminent — JPMorgan CEO Jamie Dimon has called for light-touch federal oversight rather than alarm, and analyst Gil Luria dismissed talk of an immediate threat outright, saying "there is no threat to humanity right now."
Source: Fortune
