Anthropic's CEO published an essay arguing AI is growing exponentially. His company's own researchers, writing the same week, said it isn't.
Dario Amodei's essay, titled "Policy on the AI exponential," argues that AI scaling laws guarantee continued explosive growth in general cognitive capabilities. His central citation is a 2020 paper on neural language model scaling, co-authored by Anthropic co-founder Jared Kaplan during his OpenAI tenure. But the system cards Anthropic published for its two newest frontier models say something else. The Claude Mythos card states that "early claims of large AI-attributable wins have not held up." The Fable 5 card is more specific: Anthropic researchers say they "do not observe a sustained, AI-attributable 2x acceleration in the pace of our AI progress."
The gap between executive messaging and internal findings is not a minor discrepancy. It goes to the core of how Anthropic justifies its valuation and frames its case for regulatory urgency, especially with an IPO approaching. The scaling-laws narrative is load-bearing for the investor pitch, and Amodei is leaning on a six-year-old paper that has faced sustained challenge from researchers outside the AI-booster camp.
Gary Marcus spent four years being mocked for arguing that scaling laws were hitting diminishing returns. The Fable 5 system card just handed him a footnote.