In a quotation collected by Simon Willison and posted on 9th August 2026, an excerpt from the Claude Opus 5 system prompt describes how the model should handle questions about the suspension and restoration of two earlier Claude models. According to the excerpt, Claude Fable 5 and Claude Mythos 5 were first released on June 9, 2026. On June 12, 2026, Anthropic suspended access to both models to comply with U.S. Department of Commerce export controls. The Department lifted those controls on June 30, 2026, and Anthropic restored access on July 1, 2026, per Anthropic's linked statement. The excerpt notes these events occurred after Claude's training-data cutoff, so the model knows of them only through this notice. The prompt instructs Claude to confirm the suspension accurately and 'matter-of-factly' if asked, to give 'a fair, accurate account rather than sharing personal opinions' on the export controls, and to point users to Anthropic's statement. It also tells Claude to check for newer information when it can search, and otherwise to suggest checking Anthropic's site. The excerpt is attributed to the Claude Opus 5 system prompt.
- System prompt excerpt quoted by Simon Willison, posted 9 August 2026
- Claude Fable 5 and Mythos 5 suspended June 12, 2026 for export controls; access restored July 1, 2026
What it means for you
A 'system prompt' is the hidden set of instructions a company gives its AI before you ever type anything. This excerpt shows Anthropic writing plain instructions telling Claude how to talk about a real event — a brief suspension of two older models — that happened after the model was trained. For most users, this changes nothing about how you'd use the tool; it's a window into how these systems are steered behind the scenes.
Try this
If you build on Claude, read Anthropic's published system prompts to understand what behaviour is being set for you before your own instructions apply.
Who should care
Developers, prompt engineers, and anyone curious about how AI vendors shape model behaviour on sensitive or recent topics.
Skip this if
You just use Claude to draft emails, summarise documents, or answer everyday questions — this has no effect on your work.
Sources: Simon Willison — read the original