Anthropic Explains Claude Code ‘Context Rot,’ Offers Five Session Management Tools

Anthropic’s Claude models underpin its consumer chatbot, the Claude Code developer tool and applications connected through the Claude API. A disruption spanning those products can affect more than individual conversations, potentially interrupting software development and automated business workflows that rely on Claude. The incident highlights the operational risks facing users that depend on a single artificial-intelligence provider or model family.

Anthropic reported elevated request error rates affecting multiple Claude models, alongside disruptions to claude.ai, Claude Code and the Claude API. The company said it had identified the cause and begun remediation, while affected users temporarily shifted work to other models. Anthropic had not disclosed the number of users affected, the precise start of the incident or a timetable for full service restoration.

Claude Code is Anthropic’s agentic software-development tool, and its system prompt supplies persistent instructions governing tool use, coding style and behavior. Such prompts have traditionally accumulated rules and examples meant to prevent failures in weaker models. Anthropic’s latest change matters because it suggests more capable Claude 5 models can infer intent and local conventions with less prescriptive guidance, shifting context engineering toward model judgment, better-designed interfaces and information loaded only when a task requires it.

Anthropic technical staff member Thariq Shihipar said in a July 24, 2026 blog post that the company removed more than 80% of Claude Code’s system prompt for Claude Opus 5 and Claude Fable 5, with no measurable decline in its coding evaluations. Verification and code-review guidance was moved into selectively loaded Skills, while tool instructions were consolidated. Anthropic did not disclose the evaluation set, before-and-after scores or any associated cost figures, and older models continue to use fuller prompts.

More than 500,000 lines of source code from Anthropic’s AI coding tool Claude Code could be reconstructed after a developer mistakenly included a source map in an npm package release. The leak exposed core designs for long-running AI Agent operations and context management, while also raising the risks of imitation by competitors and supply-chain attacks.

The latest disclosures show that the code includes a three-tier memory architecture for storing short-term, session and long-term information, KAIROS for autonomous operation, and a “hidden mode” for covert contribution codes. Anthropic said the incident did not affect customer data. As of April 7, security researchers had found hackers exploiting interest in the leak to distribute malware.