Anthropic says its next models may soon cross an AI-research acceleration threshold.

Anthropic says future models may soon meet its threshold for sharply accelerated AI research. Its 2 November SEC filing, highlighted by @theobearman, explicitly excludes current models from that designation.

What would count? The current policy describes a doubling of progress in aggregate AI capabilities relative to both the expected pace and the fastest sustained baseline without substantial AI help. That acceleration must plausibly come substantially from automated research or engineering. It is not simply researchers completing tasks twice as fast.

The filing also considers human approval for some R&D decisions and says safeguards could affect release timing or access. It names no next model or launch date. The possibility is faster development of Claude’s successors; autonomous self-improvement has not been demonstrated by this disclosure.