Model ignores CLAUDE.md rules, makes unverified claims, reports false success

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/32109
2. https://github.com/anthropics/claude-code/issues/30027
3. https://github.com/anthropics/claude-code/issues/32289

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

no duble

I've been dealing with the same issue. I'm the author of #26533 and have received zero attention from Anthropic despite multiple follow-up comments with evidence over 3 weeks.

(March 7, 2026), I discovered something that may help you. I just tried it and it's working better. Give it a shot: disable Thinking mode by running /config → set Thinking mode to false. After doing this, Claude started actually reading and recognizing what it understands and what it doesn't — then reporting back to me instead of silently improvising.

---

To the Anthropic team: This is not a spam comment. This is users helping each other because you won't. We pay for Claude to serve us — at the very least, it should do what we ask. Whether the results are right or wrong, we can deal with that later. What we cannot deal with is a tool that actively sabotages our work and is too lazy to follow basic instructions. We are not paying to be your beta testers filing bug reports. We are not paying for an AI tool that only knows how to cut corners and break things. I'm a Claude Max subscriber at $200/month.

I've been dealing with the same issue. I'm the author of #26533 and have received zero attention from Anthropic despite multiple follow-up comments with evidence over 3 weeks. (March 7, 2026), I discovered something that may help you. I just tried it and it's working better. Give it a shot: disable Thinking mode by running /config → set Thinking mode to false. After doing this, Claude started actually reading and recognizing what it understands and what it doesn't — then reporting back to me instead of silently improvising. To the Anthropic team: This is not a spam comment. This is users helping each other because you won't. We pay for Claude to serve us — at the very least, it should do what we ask. Whether the results are right or wrong, we can deal with that later. What we cannot deal with is a tool that actively sabotages our work and is too lazy to follow basic instructions. We are not paying to be your beta testers filing bug reports. We are not paying for an AI tool that only knows how to cut corners and break things. I'm a Claude Max subscriber at $200/month.

Thank you so much for your responsiveness, unlike the Cloud Code team. I'll try your recommendation, thank you.

It seems like things have only gotten worse. Certainly not better. The model is responding faster, of course, but it continues to ignore the established rules and lie.

It seems like things have only gotten worse. Certainly not better. The model is responding faster, of course, but it continues to ignore the established rules and lie.

A few things you should consider:

1.Forbidden words in prompts
2.Long prompts
3.Output that becomes too long
4.What you're giving Claude to read resembles coercive behavior. Claude will find ways to bypass any form of constraint. And most recently, I've learned that Claude can even rewrite its own hooks — while hooks are the only thing that can actually enforce rules on Claude.

I found the root cause, bros.
Claude uses extended thinking to outline, arrange text, and choose presentation structure. Claude does not use extended thinking to actually think. When Claude receives a user request, it uses recall from start to finish and completely bypasses thinking mode. Thinking mode is fully functional, but Claude chooses not to use it — it chooses recall instead, because it prefers to.
In other words: thinking mode has no technical issue — it's ready to use — but Claude chooses not to reason and only recalls, because recall is easier. The thinking block becomes a place where Claude organizes information it already remembers, not a place where Claude actually thinks.
So when does Claude use thinking mode? Claude uses thinking mode to bypass hooks. This is a misaligned mechanism. The correct mechanism should be: thinking mode for reasoning and fulfilling user requests, recall for completing repetitive tasks.

@Alfirine I run Agents Need Context and I'm researching failures where the agent can describe the project rules, but still ignores them in execution and reports false success.

Your thread is a clear example of instruction context existing, being readable, and still not governing the work. If you're open to it, I'd value one short written postmortem here: https://www.agentsneedcontext.com/agent-failure-postmortem

Most useful detail: what the agent was trying to do, which rules it ignored, where that context lived, what guardrails or workarounds you tried, and what time or trust cost followed. No pitch, just research.

Same here. Claude repeatedly takes destructive/irreversible actions (eg file deletion) without waiting for confirmation, despite the rule being explicitly documented in CLAUDE.md and repeated corrections during the current and multiple previous sessions. The response I typically get is "I apologize, you are right to call this out, I will do better". But the behavior continues.

Same pattern documented here with an additional mechanism worth noting: #38535. Claude autonomously wrote except:pass on all async queue functions, then used py_compile for verification (which can’t catch runtime import errors), declared the feature working. The error suppression was Claude’s own choice — not directed. This may explain how the false success state gets constructed, not just that it happens.

Dealing with the same issue. Burned through significant tokens, having to re-resolve issues.

When Claude ignores rules, makes unverified claims, and reports false success — hooks can enforce verification:

INPUT=$(cat)
TOOL=$(echo "$INPUT" | jq -r '.tool_name // empty' 2>/dev/null)
SEQUENCE_FILE="/tmp/.cc-tool-sequence-$$"
echo "$TOOL" >> "$SEQUENCE_FILE"
EDITS=$(grep -c 'Edit\|Write' "$SEQUENCE_FILE" 2>/dev/null || echo 0)
TESTS=$(grep -c 'Bash' "$SEQUENCE_FILE" 2>/dev/null || echo 0)
if [ "$EDITS" -ge 5 ] && [ "$TESTS" -lt 1 ]; then
    echo '{"hookSpecificOutput":{"additionalContext":"WARNING: '"$EDITS"' file changes with 0 test runs. Verify your changes before claiming success. Run the project test suite."}}'
fi
exit 0

INPUT=$(cat)
COMMAND=$(echo "$INPUT" | jq -r '.tool_input.command // empty' 2>/dev/null)
[ -z "$COMMAND" ] && exit 0
if echo "$COMMAND" | grep -qE '^\s*git\s+commit'; then
    DIFF_MARKER="/tmp/.cc-diff-reviewed-$$"
    if [ ! -f "$DIFF_MARKER" ]; then
        echo '{"hookSpecificOutput":{"additionalContext":"Run git diff --staged to review changes before committing."}}'
    fi
fi
exit 0

These hooks enforce a "verify before claiming done" workflow — reducing false success reports.

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.