[claude.ai] CRITICAL: Input Field Randomly Replaces User Text with Previous Prompt

Resolved 💬 4 comments Opened Dec 16, 2025 by sudoxreboot Closed Feb 14, 2026

Problem

The claude.ai input field randomly replaces user-typed content with a previous prompt, causing complete loss of work. This is a critical data loss bug.

Reproduction Steps

  1. Have a conversation with multiple back-and-forth messages
  2. Type a new, lengthy message (multiple paragraphs)
  3. Before sending, the input field content is randomly replaced with a PREVIOUS prompt from earlier in the session
  4. User's new content is completely lost
  5. No undo mechanism available

Observed Behavior

  • Happens unpredictably, more common in longer sessions
  • No warning or indication before replacement occurs
  • The replaced content appears to be from the browser's form autocomplete or some internal state management issue
  • Ctrl+Z does not recover the lost text

Expected Behavior

  • User input should NEVER be modified or replaced without explicit user action
  • Input field content should be treated as sacred
  • If any state restoration occurs, it should only happen on page load, never mid-typing

Impact

  • Data loss: Users lose carefully composed, lengthy prompts
  • Trust erosion: Users can't rely on the interface for important work
  • Workflow disruption: Forces external composition and paste-in as workaround
  • Frustration: This is a $200/mo product exhibiting basic input handling bugs

Proposed Solutions

  1. Disable browser autocomplete on the input field if that's the cause
  2. Fix state management - Don't overwrite input field from internal state while user is typing
  3. Implement local draft persistence - Save input to localStorage on every keystroke
  4. Add undo history for the input field specifically

Priority

This should be P0. It causes direct, unrecoverable data loss for paying customers.

---

Alternatively, if these fixes are not feasible in the short term, update your ToS to allow subscribers to access their own sessions programmatically. This would let users build tools that work while these issues are addressed. Paying $200/mo for a product we can't reliably use, with no workaround permitted, is not acceptable.

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗