Anthropic, the AI safety and research company behind the Claude chatbot, reversed a user interface change to its Claude Code tool on February 18, 2026, following a swift and vocal backlash from developers. The update, released as version 2.1.20 on February 16, initially obscured details about which files the AI was accessing during coding tasks, replacing specific file paths with generalized summaries such as “Read 3 files.”
The alteration, which condensed the output to minimize what Anthropic described as “noise,” required users to invoke a keyboard shortcut (Ctrl+O) to reveal previously displayed file names and line counts. This change sparked immediate criticism from developers who argued that access to this information is critical for security, debugging and cost management.
“Why is the visibility of files critical?” one developer wrote on GitHub, as reported by Developpez.com. Concerns centered on the ability to quickly identify potential security vulnerabilities, trace the AI’s actions, and monitor token consumption – a significant cost for users of Claude Code, with subscriptions reaching up to $200 per month. Without knowing precisely which files Claude Code was reading, writing, or modifying, developers expressed difficulty in auditing the AI’s activity and ensuring it was operating within secure parameters.
The initial response from Anthropic framed the change as a simplification intended to allow developers to focus on code diffs, and outputs. But, the developer community overwhelmingly rejected this rationale. One post characterized the update as “dumbing down” the tool, emphasizing the practical difficulties introduced by the obscured file access details.
Anthropic addressed some of the concerns in a subsequent release, version 2.1.47, published on February 18, 2026. The release notes, posted on GitHub, detail fixes for file write tool line counting, Windows terminal rendering bugs, and improvements to VS Code plan preview functionality. The notes do not directly address the file access visibility issue, but indicate a broader effort to improve the tool’s performance and usability.
The incident highlights a growing tension in the design of agentic AI coding tools: determining the appropriate level of abstraction between the AI’s actions and the developer’s oversight. As these tools grow more integrated into the software development lifecycle, the need for transparency and control over their operations is becoming increasingly apparent. Anthropic has not publicly commented on the specific reasons for the initial change or its decision to revert it.
