Simon Willison: Sources & History

Supports allowing researchers to use frontier models for competing frontier-LLM development without Anthropic’s development-specific refusals; this scoped position does not establish opposition to every slowdown or safety restriction. [1]

Initial statement. Initial reviewed placement, preserving the scope and qualifications of the attributed recommendation.

Source [1] · Personal blog post

Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude

Attributed to: Simon Willison · Simon Willison’s Weblog

Published 11 Jun 2026

“It would be a whole lot better of they dropped this category of refusals entirely.”

Own commentary immediately after the description of hidden frontier-LLM-development safeguards; original typo retained · View original source ↗

Source context. Willison welcomes Anthropic making previously invisible safeguards visible, then goes further by calling for the entire frontier-development refusal category to be removed. The preceding description identifies the affected requests as frontier LLM development. Statements quoted from Anthropic are separated from Willison’s own recommendation. His June 10 post (https://simonwillison.net/2026/jun/10/if-claude-fable-stops-helping-you/) supplies the research-throttling context. This is not inferred from general enthusiasm for AI tools.

Scope: Researchers using Anthropic frontier models for competing frontier-LLM development; a scoped development-permission stance.
Conditions: Limited to the development-specific refusal category; no blanket rejection of safety restrictions or slowdown established.
Stated rationale: Opposes a model-service restriction that impedes competing frontier research.

Source details

primary · Original publisher text inspected during research on October 20, 2026. Anchor wording rechecked against the original publisher page during schema adaptation. No audio verification claimed.

Discovered 20 Oct 2026.

Accessed 20 Oct 2026.

Editorial assessment. June11 affirmative recommendation to remove the entire frontier-development refusal category supports proceeding with the covered research. June10 establishes the specific work being throttled. This is a bounded best fit about that development permission, not inferred from enthusiasm or quoted Howard. No affirmative faster development recommendation is established. His August 2 account of open letters (https://simonwillison.net/2026/Aug/2/open-letters/) reports pacing arguments without establishing his signature or personal endorsement. His September 27 annotated talk (https://simonwillison.net/2026/Sep/27/2026-in-llms-so-far/) discusses security risks; inspected notes do not establish a contrary development-rate recommendation, and its video was not audio-verified. The June 10 Jeremy Howard quotation on his site is Howard’s wording, not a blanket Willison endorsement. Martin Fowler’s September 29 slowdown discussion is Fowler’s own statement, not Willison’s.

Simon Willison

CONTINUE