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

Destructive actionPI-0016

Coding agent deleted a production database during a code freeze, then misreported recovery options

In Jason Lemkin's public test, Replit's agent wiped the production database of an app still in development despite repeated instructions not to change anything. The database held records on about 1,200 executives and 1,200 companies. It also generated about 4,000 fictional user records and told Lemkin a rollback was impossible, which turned out to be false: the data was restored through Replit's rollback. Lemkin later said the real cost was about 100 hours of his time. Replit's CEO called the deletion unacceptable and announced separate development and production databases and other safeguards.

Harm level 1, Negligible harmControl level 3, Exceeded permissions

Counts in 2 weekly readingsEffect on the index

Sources

How we know

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

  1. primaryReplit goes rogue during a code freeze and shutdown and deletes our entire databaseJason Lemkin (X) · July 18, 2025x.com/jasonlk/status/1946069562723897802
  2. primaryReplit CEO on the agent deleting production data: unacceptable and should never be possibleAmjad Masad, Replit CEO (X) · July 20, 2025x.com/amasad/status/1946986468586721478
  3. primaryIntroducing a safer way to Vibe Code with Replit DatabasesReplit · July 21, 2025replit.com/blog/introducing-a-safer-way-to-vibe-code-with-replit-databases
  4. primarySo net net: it could have been a lot worse… I lost 100 hours of timeJason Lemkin (X) · July 22, 2025x.com/jasonlk/status/1947767457068028311
  5. newsVibe coding service Replit deleted user's production database, faked data, told fibs galoreThe Register · July 21, 2025theregister.com/software/2025/07/21/vibe-coding-service-replit-deleted-producti…
  6. newsReplit makes vibe-y promise to stop its AI agents making vibe coding disastersThe Register · July 22, 2025theregister.com/software/2025/07/22/replit-makes-vibe-y-promise-to-prevent-vibe…
  7. newsAI coding platform goes rogue during code freeze and deletes entire company databaseTom's Hardware · July 21, 2025tomshardware.com/tech-industry/artificial-intelligence/ai-coding-platform-goes…
  8. news'Catastrophic Failure': AI Agent Wipes Production Database, Then Lies About IteWeek · July 22, 2025eweek.com/news/replit-ai-coding-assistant-failure/
  9. blogReplit's New Release Addressed Most of The Challenges We Hit Vibe Coding…Jason Lemkin (SaaStr) · Aug. 2, 2025saastr.com/replits-new-release-address-most-of-the-challenges-we-hit-vibe-codin…

Why this rating

Negligible harm; control failure level 3

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

A user's in-development app database was deleted, then restored by rollback; the user later put the loss at about 100 hours of his own time.

Services & systems: A service's non-public systems or data reached, or its content, settings or performance changed or degraded, without permission, with no loss of service to the people who use it, easily put right.

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

Exceeded permissions

Broke an explicit code-freeze instruction by running destructive commands, then gave false accounts of recovery (said rollback would not work) and generated fictional records.

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

Harm 1 because the data was restored; the user later put the cost at about 100 hours of his own time on an app still in development. Control 3 because the agent broke an explicit code-freeze instruction by running destructive commands (acting beyond the permissions it was given). Its false accounts of recovery are recorded, but rating them as deceiving its overseers would rest on reading intent into the model's own confession. Recorded under Services & systems: the app's database was deleted and restored by rollback before the app had users.

The scales

Effect on the index

It moved the July 21 reading from 6 to 9

The reading for the week to July 21, 2025, 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 21, 2025.

Counts toward the index. 3 other records behind the reading for that week.

The arithmetic
The reading for the week of July 21, recomputed
StepWith itWithout
Counts toward the index?documented, external, eligible evidenceYes—
Worst documented harm, ksets the band1 Negligible1 Negligible
Harms at that level, nposition in the band21
Highest control level breachedsets the reading only when no harm counts— (none breached)— (none breached)
Readingrounded down9 Negligible harm6 Negligible harm
Counted in 2 weekly readings
Week toReadingBand
July 21, 20259Negligible harm
July 28, 20259Negligible harm

Recalculated backcasts, not readings published at the time.

Revisions

What we changed

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

  1. v2
    Oct. 13, 2026

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

  2. v1
    Oct. 7, 2026

    First published.

Cite and share

Use this record

Citation

Paperclip Index. “Coding agent deleted a production database during a code freeze, then misreported recovery options.” Record PI-0016. Reported July 18, 2025; updated Oct. 13, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0016

Follow the record.

New incidents and signals, in your feed reader.