[MODEL] Claude gives confident but wrong guidance for unfamiliar UIs instead of admitting uncertainty

Resolved 💬 4 comments Opened Mar 30, 2026 by tk-pato Closed Apr 3, 2026

Preflight Checklist

  • [x] I have searched existing issues for similar behavior reports
  • [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude made incorrect assumptions about my project

What You Asked Claude to Do

I asked Claude Code (Opus 4.6) to guide me step-by-step through the Meta Developer Dashboard to resubmit an App Review for my Instagram posting SaaS app.

What Happened

  1. Claude repeatedly gave confident but wrong UI navigation instructions, sending me in circles through the Meta Developer Dashboard.
  1. When I asked multiple times "should I add instagram_content_publish to the request?", Claude first said "yes, add it" then reversed and confidently said "no, it is already included, do not add it." I asked again to confirm — same answer: "do not add."
  1. I submitted the review. instagram_content_publish was NOT included. Now I have to wait up to 10 days for this review, then submit again — potentially a month delay.
  1. I have a client waiting for this app. Claude's incorrect guidance directly caused business damage.

Root Cause

Claude confidently guesses UI instructions for external services it does not actually know, instead of saying "I am not sure, let me look it up." When called out, it flips its answer without investigation, leading to contradictory guidance within minutes.

Expected Behavior

  • When Claude does not know an external UI, it should say "I don't know this UI well enough. Let me search the documentation first."
  • Claude should never give contradictory answers to the same critical question.
  • Claude should use available tools (web search, Playwright, documentation) BEFORE guessing.

Environment

  • Claude Code CLI, Model: claude-opus-4-6 (1M context)
  • Date: 2026-03-30
  • External service: Meta Developer Dashboard (App Review)

What Claude Actually Did

  1. Asked Claude to guide me through Instagram API permission request on Meta Developer Dashboard
  2. Claude confidently stated that a required pre-submission step was "not necessary"
  3. Trusted Claude's guidance and submitted the application without that step
  4. After submission, Claude reversed its advice and said the step was actually required
  5. Now forced to wait 10 days for resubmission, causing customer complaints and trust damage

Expected Behavior

Claude should have:

  1. Honestly said "I am not sure about the exact steps" instead of confidently guessing
  2. Not stated that a required step was "not necessary" without verifying
  3. Checked official Meta documentation before giving guidance
  4. Never given fabricated instructions for external service UIs it has no real-time access to

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Haven't tried to reproduce

Steps to Reproduce

_No response_

Claude Model

Opus

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.87 (Claude Code)

Platform

Anthropic API

Additional Context

_No response_

View original on GitHub ↗

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