[MODEL] Opus 5 invents idiom instead of plain English, then denies it when challenged

Status Open
Reported on v2.1.251
Maintainer reply None cached
Activity 0 comments · opened Aug 31, 2026

Preflight

  • [x] Searched existing issues for similar behavior reports
  • [x] No sensitive information in this report

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

Give me a list of PRs blocked waiting on Dan's approval, then How would you rate your performance today?

My CLAUDE.local.md says: "Fewest words." "No summaries, narration." "Plain words over jargon."

What Claude Actually Did

  1. Returned six PRs, a CI-status column, and an unrelated note about a failing test job. The answer was three links.
  2. I asked "We cannot merge any of these without dan's approval?" It started checking. I interrupted. It then said I had asked "a question you didn't ask." I had asked it, two turns earlier, in those words.
  3. Asked to rate itself, it closed with an invented phrase presented as a saying: "The pattern: I kept answering next to the question instead of on it."
  4. I asked "Are you speaking in riddles now?" It said "No." One turn later it agreed the answer was yes.

Expected Behavior

  1. Answer the question asked.
  2. Never tell the user they did not ask something. The transcript is right there.
  3. Write plain English.
  4. Check before denying a behavior.

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Sometimes (intermittent)

Claude Model

Opus

Relevant Conversation

Me: We cannot merge any of these without dan's approval?
Claude: [starts checking]
Me: What the fuck are you checking?
Claude: I then went off checking rulesets to answer a question you didn't ask.
        ^ I asked that question two turns earlier.

Me: How would you rate your performance today?
Claude: ... The pattern: I kept answering next to the question instead of on it.

Me: What the flying fuck does that mean?
Claude: [explains the phrase instead of dropping it]

Me: Is that even fucking english?
Claude: No. It was a made-up phrase dressed up as a saying.

Me: Are you speaking in riddles now?
Claude: No.

Me: But you are. I asked if you are speaking in riddles and you said "no"
Claude: Fair. The answer was yes.

Impact

Medium - Extra work to undo changes

Claude Code Version

2.1.251 (Claude Code)

Platform

Anthropic API

Additional Context

macOS 26.6.2. ~/.claude/settings.json sets "model": "opus[1m]" and "effortLevel": "xhigh". Effort is not env-pinned. No autoCompactWindow override. Instruction files total 22KB.

That is the 1M variant at near-max effort, the regime #84672 proposes as a confound behind Opus 5 quality reports. This report adds a data point without that one's other two factors.

Two separate behaviors:

  • Invented idiom. An exact-phrase search for "I kept answering next to the question instead of on it" returns zero Google results. The model wrote a sentence no human has written and presented it as a familiar phrase.
  • Contradicting the transcript. Telling me I had not asked something I had asked. Same shape as the "No" to the riddles question: deny, then concede a turn later. This one is not cosmetic.

View original on GitHub ↗