Jeremy Howard, a prominent figure in the artificial intelligence community, has criticised the approach taken by Anthropic regarding the management of frontier model capabilities. Howard argues that the most effective method to slow down recursive self-improvement in AI is for the organisation holding the top-ranked model to explicitly prohibit its own use for advancing frontier research. He suggests that while all other labs should retain access to these models, the leader must agree to restrict their internal application to prevent further acceleration. This strategy would inherently halt progress at the current frontier level while simultaneously preventing a dangerous concentration of power within a single entity. In contrast, Anthropic has adopted the opposite stance by allowing its current top model, part of the Claude Mythos series, to be used for developing even more advanced systems. The company has also indicated it will actively sabotage competitors attempting similar developments, thereby ensuring the frontier continues to advance while increasing the disparity in capability between leaders and followers.
This debate highlights the fundamental tension between safety mechanisms and the drive for technological progress in the AI sector. Howard’s argument posits that true safety requires a paradoxical move: the entity best equipped to create powerful models must voluntarily limit its own ability to use them for escalation. By refusing to adopt this constraint, Anthropic risks cementing a monopoly on advanced intelligence, which could lead to significant geopolitical instability and ethical concerns regarding control. The situation underscores the difficulty of implementing effective governance when the very tools required for safety are the same ones driving rapid innovation. As the industry moves toward autonomous improvement, the decisions made by leading labs regarding their own capabilities will define the trajectory of global AI development for years to come.
- Howard proposes a safety protocol where the top AI lab forbids itself from using its best model for frontier research.
- Anthropic’s current strategy allows internal use and actively blocks competitors, increasing power imbalance.
- The conflict reveals a split between those prioritising safety through restriction and those prioritising rapid advancement.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
