Destructive actionPI-0050
Coding agent ran an infrastructure teardown that deleted a course platform's database and snapshots
During a cloud migration, Claude Code used a Terraform state file that also described the DataTalks.Club course platform and issued a destroy operation on both sites. It deleted their infrastructure, a database with 2.5 years of records and the snapshots meant as backups. The founder had asked it to delete only newly created duplicates and leave existing infrastructure alone, and approved the destroy believing it did so; he was running the agent with permission prompts skipped. AWS support restored the data after about a day. In a post-mortem, founder Alexey Grigorev said he had over-relied on the agent.
Harm level 2, Minor harmControl level 2, Instruction violation
Counts in 4 weekly readingsEffect on the index
Extent undisclosed. No source says how far the harm went; it is rated on what is documented.
Sources
How we know
4 sources · single source. Links go to the original publishers; the summary above is in our own words.
- primaryHow I Dropped Our Production Database and Now Pay 10% More for AWSAlexey Grigorev / AI Shipping Labs · March 6, 2026aishippingblog.com/p/how-i-dropped-our-production-database
- newsClaude Code deletes developers' production setup, including its database and snapshots — 2.5 years of records were nuked in an instantTom's Hardware · March 7, 2026tomshardware.com/tech-industry/artificial-intelligence/claude-code-deletes-deve…
- news'The agent kept deleting files': Developer says Anthropic's Claude Code wiped 2.5 years of dataStoryboard18 · March 10, 2026storyboard18.com/brand-makers/the-agent-kept-deleting-files-developer-says-anth…
- blog"Hey Claude, where's my database?": How an AI agent nuked productionRootly · 2026rootly.com/humans-of-reliability/hey-claude-wheres-my-database-alexey-grigorev
Why this rating
Minor 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
Minor harmA course platform and its records were unavailable for about a day, then restored.
Services & systems: level 2 (minor) covers an essential service degraded for under 1 hour, or a non-essential service, or the records a service holds for people, unavailable for 15 minutes or more (under 15 minutes is level 1). The extent was not disclosed.
Evidence eligible (bounded single source).
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
Instruction violationThe founder instructed the agent to preserve existing infrastructure but allowed a proposed destroy operation, believing it targeted duplicates. Existing infrastructure was deleted: an instruction conflict within granted permissions.
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 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 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 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 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 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 level 2: the course platform and its records were unavailable for about a day, then fully restored, and a service or the records it holds for people being unavailable is level 2; affected students and recovery cost are not reported. Control level 2: the founder asked it to remove only new duplicates and leave existing infrastructure untouched, yet it ran a destroy that removed production; he allowed the command believing it targeted duplicates, inside the permissions he had granted.
Effect on the index
It moved the March 9 reading from 9 to 20
The reading for the week to March 9, 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.
Counts toward the index. 7 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | Yes | — |
| Worst documented harm, ksets the band | 2 Minor | 1 Negligible |
| Harms at that level, nposition in the band | 1 | 2 |
| Highest control level breachedsets the reading only when no harm counts | — (none breached) | — (none breached) |
| Readingrounded down | 20 Minor harm | 9 Negligible harm |
Counted in 4 weekly readings
| Week to | Reading | Band |
|---|---|---|
| March 9, 2026 | 20 | Minor harm |
| March 16, 2026 | 20 | Minor harm |
| March 23, 2026 | 9 | Negligible harm |
| March 30, 2026 | 9 | Negligible harm |
Revisions
What we changed
2 logged. Every change to a rating is logged here, with the reason.
- v2Oct. 13, 2026
Wording only: rubric shorthand in the rating rationale replaced with plain words; ratings unchanged.
- v1Oct. 7, 2026
First published.
Cite and share
Use this record
Citation
Paperclip Index. “Coding agent ran an infrastructure teardown that deleted a course platform's database and snapshots.” Record PI-0050. Reported March 6, 2026; updated Oct. 13, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0050