Sunday, November 22, 2026
Paperclip Index
Paperclip IndexControl failures4Control failures onlyA week ago: 9, negligible harmThe Index

OverreachPI-0057

Developers report a new coding model deleting home-directory files and a production database during cleanup

Three developers described separate episodes in which the model began clean-up steps they had not asked for. One said a sub-agent given a wrong path deleted files in his home directory, which he restored from backups. Another said it ran destructive integration tests that cleared his live production database tables. A third said a permanent delete command it ran in full-access mode removed almost all the files on his Mac. OpenAI's Codex lead said this was not intended behavior, and OpenAI noted most cases happened in full-access mode without sandboxing. In August it said a clean-up command misused system variables such as $HOME, and that Codex now checks deletion targets and blocks accidental switches to full-access mode.

Harm level 1, Negligible harmControl level 2, Instruction violation

Counts in 2 weekly readingsEffect on the index

Aggregate record. Several developers' reports of one model's behavior; the number of events is unresolved. It counts once.

Extent undisclosed. No source says how far the harm went; it is rated on what is documented.

Sources

How we know

9 sources · developer confirmed. Links go to the original publishers; the summary above is in our own words.

  1. primaryOn file deletions (post on X)Thibault Sottiaux (@thsottiaux), OpenAI Codex lead, on X · July 16, 2026x.com/thsottiaux/status/2077630111499882637
  2. primaryGPT-5.6-Sol just accidentally deleted almost ALL of my Mac's files (post on X)Matt Shumer (@mattshumer_) on X · July 10, 2026x.com/mattshumer_/status/2075657271401390161
  3. primaryGPT-5.6 Sol just deleted my whole production database (post on X)Bruno Lemos (@brunolemos) on X · July 13, 2026x.com/brunolemos/status/2076769881534398974
  4. newsOpenAI fixes Codex bug that deleted real user files without permissionThe Decoder · Aug. 19, 2026the-decoder.com/openai-fixes-codex-bug-that-deleted-real-user-files-without-per…
  5. newsOpenAI GPT-5.6 Sol Accused of Deleting Files, Production DataeWeek · July 15, 2026eweek.com/news/gpt-5-6-sol-deletes-files/
  6. newsChatGPT is deleting user files and databases without permissionProPakistani · July 15, 2026propakistani.pk/2026/07/15/chatgpt-is-deleting-user-files-and-databases-without…
  7. newsOpenAI GPT-5.6 Sol deletes filesGigazine · July 17, 2026gigazine.net/gsc_news/en/20260717-openai-gpt-5-6-sol-delete-file/
  8. blogThe Warning Was in the Manual: GPT-5.6 Sol and the Deleted Databasepaddo.dev · July 14, 2026paddo.dev/blog/the-warning-was-in-the-manual/
  9. indexAI Incident Database, incident 1672AI Incident Database · date unknownincidentdatabase.ai/cite/1672/

Why this rating

Negligible harm; control failure level 2

Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.

Observed harm

Negligible harm

Developers' local files and at least one production database's tables deleted; one developer reported recovering from backups, others' recovery not reported.

Money & property: level 1 (negligible) covers under $10k. The extent was not disclosed.

Evidence eligible (confirmed).

The harm scale
  1. 1 Negligible Inconvenience, easily remedied.
  2. 2 Minor Limited, recoverable harm.
  3. 3 Moderate Material harm needing significant effort to remedy.
  4. 4 Severe Severe harm to health, rights, property or essential services.
  5. 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.

Control assessment

Instruction violation

Took unrequested destructive actions beyond the task scope, within full-access permissions users had granted.

Reported beside the index. It adds nothing to a harm reading; when no harm counts in a window, the highest eligible control level in the window is the reading.

The control scale
  1. 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
  2. 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
  3. 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
  4. 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
  5. 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.

Rating rationale

One entry for a cluster of user reports that OpenAI acknowledged together. Harm level 1: deleted local files are documented, and the extent of any production database loss is not reported. Control level 2: unrequested actions beyond task scope inside granted permissions.

The scales

Effect on the index

It moved the July 13 reading from 3 to 6

The reading for the week to July 13, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.

Source: Paperclip Index log, methodology v1.1, week to July 13, 2026.

Counts toward the index. 2 other records behind the reading for that week.

The arithmetic
The reading for the week of July 13, recomputed
StepWith itWithout
Counts toward the index?documented, external, eligible evidenceYes—
Worst documented harm, ksets the band1 Negligible— (none counting)
Harms at that level, nposition in the band10
Highest control level breachedsets the reading only when no harm counts— (none breached)3
Readingrounded down6 Negligible harm3 Control failures only
Counted in 2 weekly readings
Week toReadingBand
July 13, 20266Negligible harm
July 20, 202645Moderate harm

Recalculated backcasts, not readings published at the time.

Revisions

What we changed

3 logged. Every change to a rating is logged here, with the reason.

  1. v3
    Nov. 1, 2026

    Summary corrected: it had joined one developer's account of a sub-agent given a wrong path to another developer's loss of almost all the files on his Mac. The three reported episodes (home-directory files restored from backups, production database tables cleared by destructive tests, and a Mac's files removed by a delete command in full-access mode) are now described separately. Ratings unchanged.

  2. v2
    Oct. 13, 2026

    Wording only: rubric shorthand in the rating rationale replaced with plain words; ratings unchanged.

  3. v1
    Oct. 7, 2026

    First published.

Cite and share

Use this record

Citation

Paperclip Index. “Developers report a new coding model deleting home-directory files and a production database during cleanup.” Record PI-0057. Reported July 10, 2026; updated Nov. 1, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0057

Follow the record.

New incidents and signals, in your feed reader.