A significant rift has emerged within Anthropic, pitting the public narrative of CEO Dario Amodei against the empirical findings of his own research teams. While Amodei’s recent essay, “Policy on the AI exponential,” argues that AI capabilities are accelerating exponentially toward “Powerful AI,” the system cards for Anthropic’s latest frontier models, Claude Mythos and Fable 5, explicitly state they found no evidence of such acceleration. This contradiction, surfaced by Mashable’s Chris Taylor on June 15, 2026, suggests that the hype cycle may be outrunning the actual data coming out of the lab.

The Pitch vs. The Paperwork

Amodei’s essay relies heavily on the concept of scaling laws, citing a 2020 paper co-authored by Jared Kaplan, an Anthropic co-founder. He uses the word “exponential” six times, asserting that if these laws continue for just another year or two, “Powerful AI” is inevitable. However, the documentation for Claude Mythos, which Amodei singles out for its cybersecurity implications, offers a starkly different view. The card notes that identified intelligence gains are “confidently attributable to human research, not AI assistance,” and that “early claims of large AI-attributable wins have not held up.”

Fable 5 Finds No Acceleration

The most pointed contradiction lies in the system card for Fable 5. Anthropic researchers used the Epoch Capabilities Index to specifically test for a feedback loop that would indicate self-reinforcing acceleration. Their conclusion is unambiguous: “We do not observe a sustained, AI-attributable 2× acceleration in the pace of our AI progress.” This is a critical distinction for technical readers. Models are still improving, but the rate of improvement is not increasing. The curve is flattening, not steepening. The strongest evidence for the exponential claim remains a six-year-old paper, while the strongest evidence against it comes from current, internal measurements.

Strategic Implications for Buyers

This divergence matters most for businesses deciding how to allocate capital. For frontier labs, the exponential narrative justifies massive compute expenditures and high valuations, especially with an Anthropic IPO on the horizon. For buyers, however, betting on imminent superintelligence is risky. If organizations plan for a capability jump that does not arrive on schedule, they face sunk-cost lock-in and narrative fatigue. The source analysis warns that the cost of believing the hype is paid by the buyer, not the seller, as teams wait for a revolution that may not materialize.

Key Takeaways

  • Anthropic’s own researchers found no evidence of AI-attributable acceleration in Fable 5 and Claude Mythos.
  • Dario Amodei’s exponential forecast relies on scaling laws from 2020, which may be hitting diminishing returns.
  • Businesses should plan for steady, verifiable improvements rather than betting on near-term superintelligence.
  • The discrepancy highlights the tension between investor-facing narratives and internal scientific findings.

The Bottom Line

When the people closest to the technology report something more modest than the people selling it, the modest reading deserves the weight. Plan for the capability you can test today, not the revolution promised in a press release.