Reviewed change intelligence

What changed in AI agent instructions?

This selected evidence feed explains meaningful changes only when two attributable snapshots and a human-reviewed semantic ChangeSet already exist. It is not a universal vendor changelog, automated news feed, or claim that products without a public record remained unchanged.

Selected reviewed changes

Meaningful instruction changes with evidence

Ordered by the established publication date of the compared after snapshot, newest first.

Prompt context boundary Claude Jan 18, 2026

Product context and knowledge boundary expanded

Anthropic's highlighted January changes add named product surfaces and sharpen the instruction not to infer unlisted product details that may have changed.
Practical relevance: Revalidate product-context and support-routing assumptions when moving guidance or workflows between the compared versions. The reviewed prompt evidence supports that check, but it does not establish a runtime-performance effect.

Before → After Claude Opus 4.5 · November 24, 2025 Claude Opus 4.5 · January 18, 2026 prompt change Read the reviewed interpretation

Evidence boundary

A reviewed evidence layer, not a release-note mirror

Factual summaries resolve from ProductPackage ChangeSets, snapshots, sections, and sources. Editorial implications are labeled separately and cannot widen the established product, surface, model, or version scope.

When no eligible ChangeSet exists, no public change record is generated. That is a coverage limitation, not evidence that a product did not change.

Read the evidence and publication methodology

Products represented by reviewed changes

ClaudeAnthropic publishes dated system-prompt history for claude.ai and Claude mobile apps, separate from Claude API behavior.

Meaningful instruction changes with evidence

Product context and knowledge boundary expanded

A reviewed evidence layer, not a release-note mirror

Products represented by reviewed changes