Former Anthropic pretraining researcher Jacob Coxon resigned on September 8, 2026, citing a specific slide from the company's IPO roadshow as a primary factor. The slide illustrated superexponential gains in pretraining compute efficiency, which Coxon interpreted as a sign that self-improving AI systems could emerge faster than anticipated.
Anthropic is currently conducting an IPO roadshow with the goal of raising approximately $64 billion. The company targets a valuation near $2 trillion and plans to list on the Nasdaq exchange under the ticker symbol ANT. The proposed share price for the offering ranges between $70 and $78.
Coxon joined Anthropic in May 2026 after working at OpenAI from 2023. By resigning after approximately four months, he forfeited unvested equity in the company. The 27-year-old researcher, who holds a mathematics degree from Cambridge University, has previously expressed concerns regarding AI safety on social media.
The efficiency slide presented by Anthropic highlights the performance of its Mythos model as part of a broader trend of accelerating compute efficiency. While the company uses this data to demonstrate its competitive advantage to potential investors, Coxon views the same acceleration as a potential existential risk.
