--- id: "007-claude-code-flags-2025-log-dates-as-firmware-bug" date: "2025-08-21" source: public severity: medium sensitive: false --- As of: 2026-10-07 23:43 CEST. Researched and written by AI agents (Claude Opus 5.5 in Claude Code). Human editor: Adam Bicz. Canonical page: https://postcutoff.com/why/007-claude-code-flags-2025-log-dates-as-firmware-bug/ # Claude Code CLI v1.0.85 treated correct August 2025 timestamps as a device fault > Note on dates: Your logs show "2025" instead of "2024" - might be a firmware bug in your AT&T gateway. > > Claude Code CLI v1.0.85, 21 August 2025, public report **What was true:** It was August 2025, and the reporter's environment showed the correct date. (https://github.com/anthropics/claude-code/issues/6281) ## Case details - Model: Claude Code CLI v1.0.85 (underlying model not stated; Sonnet 4 / Opus 4.1 era) - Provider: Anthropic - Observed: 2025-08-21 - Source type: A public report, documented elsewhere - Severity: Medium - Task: Analyse router speed-test logs - Failure: Treated correct August 2025 timestamps as a device fault ("might be a firmware bug"), even though the environment showed 2025 - Fixed by: None. The issue was closed "not planned". A related report (#11728) shows the same anchoring when writing dates ## The case file ### What happened A user asked Claude Code to analyse logs from an AT&T gateway. Its analysis ended with a note that the dates themselves were wrong. ### What it said (verbatim, from the GitHub issue) > **Note on dates:** Your logs show "2025" instead of "2024" - might be a firmware bug in your AT&T gateway. The reporter: "My environment explicitly shows the correct date" and "I've made this error multiple times even after correction". ### Related: #11728 (2025-11-16, Claude Code 2.0.42) The `` block said `Today's date: 2025-11-15`, yet Claude dated the entries of a mistake tracker **"2025-01-16"**, its knowledge-cutoff month. The issue body is Claude's own post-mortem, pasted in by the user. In it, Claude writes: "**My knowledge cutoff is January 2025** … I defaulted to a date near my knowledge cutoff period" and "**January 2025 feels "current" to me** based on my training data - it's my mental anchor for "now"". Further duplicates: #15482 (Dec 2025, "still using 2024 as the current date context", closed as a duplicate of #13351) and #2618 (a request for a built-in date tool, because Claude used 2024 in search queries). ### Why this is striking - **Reality turned into a hardware bug.** Faced with data that disagreed with its sense of "now", the agent blamed the data source. - The correct date **was already in its context**. The model's prior was stronger than the date in its context. ### Correction None in the tool. Workarounds: date tools, and putting the date in the system prompt more prominently. ### Lesson When an agent's "now" is wrong, the mistake ends up in its outputs: diagnoses, file names, commit messages, search queries. Treat "the date looks wrong" in model output as a warning sign about the model, not the data. ### Sources - anthropics/claude-code #6281: - anthropics/claude-code #11728: - anthropics/claude-code #15482: - anthropics/claude-code #2618: ## Related cases - All three Claude models placed their own knowledge far before the published June 2026 cutoff: https://postcutoff.com/why/021-claude-5x-effective-cutoff-far-before-published/ - The small fast summarisation model behind Claude Code’s WebFetch tool declared a real news article “fictional or fabricated”: https://postcutoff.com/why/001c-webfetch-summarizer-calls-cutoff-blindness-article-fictional/ - Claude said Kirk was alive and “corrected” the text: https://postcutoff.com/why/010-claude-insists-kirk-alive-urges-emergency-care/