{"meta":{"exportedAt":"2026-10-25T11:39:20.528Z","formatVersion":2,"selection":{"q":"critical","system":"","harm":"","context":"","country":"","role":"","relation":"","evidence":"","year":"unknown","response":"","severity":"","verification":"","view":"incidents","sort":"added"},"totalIncidents":2,"coverage":{"cases":2,"countries":1,"languages":1,"unknownLocation":1,"locationPending":0,"unknownLanguage":0,"unknownDate":2,"lawsuits":0,"regulatory":0,"minors":0,"coreRelations":1,"contextualRelations":1,"mixedRelations":0,"unknownRelations":0,"relationPending":0,"relationUnknown":0},"countingNote":"Distinct public cases in this selection. People counts apply within individual cases only; cross-case person overlap has not been resolved. No population incidence estimate.","affectedCountNote":"Interpret person counts with affectedCountStatus and the reported effects. Unquantified zeros are placeholders, not a measured zero.","source":"AI incidents","publisher":"NOPE","url":"https://nope.net/incidents","license":"CC BY 4.0"},"incidents":[{"id":"2026-us-hospital-nurses-say-ai-scheduling-tool-overrode-shift-requests-and-assigned-days-off","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'Since 2023, HCA has rolled out an AI scheduling tool it co-developed with Palantir called Timpani at roughly 130 of its 190 locations.'; 'it uses algorithmic forecasts of patient volumes and staffing needs to automatically build nursing schedules'; 'The efficiency comes from AI generating stronger initial schedules,'","relation":"supports","source_id":"s1"}],"assertion":"HCA co-developed Timpani with Palantir as an AI scheduling tool, has rolled it out at roughly 130 of 190 locations since 2023, and uses it to build nursing schedules automatically from algorithmic forecasts; an HCA executive wrote that its efficiency comes from AI generating stronger initial schedules.","causal_attribution":"WIRED's description and the executive's paper as quoted; this establishes the AI scheduling role."},{"id":"c2","status":"reported","evidence":[{"locator":"'Over the past four months, Amber Retzloff, a critical care nurse in Florida, requested to work 50 specific 12-hour shifts. But over half the time, she says her hospital'; 'Palantir-powered scheduling software assigned her to different shifts, often putting her on back-to-back-to-back days and leaving her mentally drained.'","relation":"supports","source_id":"s1"}],"assertion":"A Florida critical care nurse said that over four months the scheduling software assigned her different shifts for over half of the 50 she requested, often back-to-back days that left her mentally drained.","causal_attribution":"The nurse's own account to WIRED."},{"id":"c3","status":"reported","evidence":[{"locator":"'Barker says now essentially every nurse has been assigned to work on a day they were supposed to be off. Barker delayed a long-planned doctor'; 'appointment for a chronic condition after struggling to trade an assigned red day.'","relation":"supports","source_id":"s1"}],"assertion":"A Missouri nurse said the tool assigned essentially every nurse at the facility to work a requested day off, and that the nurse delayed a long-planned doctor's appointment for a chronic condition after struggling to trade such a day.","causal_attribution":"The nurse's own account to WIRED."},{"id":"c4","status":"reported","evidence":[{"locator":"'says that nursing leaders, not Timpani, make final scheduling decisions'; 'On average, Timpani schedules nurses for 1 percent of their requested days off, or what are known as red days, according to HCA.'","relation":"supports","source_id":"s1"}],"assertion":"HCA says nursing leaders, not Timpani, make final scheduling decisions, that the tool schedules nurses on 1 percent of requested days off on average, and that it keeps improving the tool on nurses' feedback.","causal_attribution":"HCA's statements, preserved as counterevidence on who decides."}],"effects":[{"label":"nurses say an AI scheduling tool overrode their shift requests and days off, leaving one exhausted and leading another to delay a medical appointment","claim_id":"c3","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.wired.com/story/ai-making-mess-of-nurses-schedules-they-say-its-a-safety-issue/","kind":"news_report","access":"read","language":"en","translation_note":"Read in full in English on 2026-10-13 (HTTP 200); no translation. WIRED reports interviews with nurses, HCA's and Palantir's statements and an HCA executive's paper.","independence_group":"wired-nurse-interviews"}],"version":1,"ai_roles":["institutional_use"],"contexts":["work","health"],"unknowns":["The dates of the specific assignments described; WIRED reports the four months before its October 2026 article for the Florida nurse.","Whether managers reviewed or edited the specific schedules the nurses describe.","The outcome of the Missouri union grievance."],"geography":{"basis":"WIRED places the nurses at HCA hospitals in Florida and Missouri, United States.","court_countries":[],"event_countries":["US"],"affected_person_countries":["US"]},"publication":{"basis":"Published under the approved AI workforce-scheduling class: two named nurses told WIRED that HCA's AI scheduling tool, which builds schedules automatically, assigned them shifts and a requested day off against their preferences, with exhaustion and a delayed medical appointment as consequences. HCA disputes that the tool makes final decisions.","reviewed_on":"2026-10-13"},"ai_involvement":{"basis":"WIRED describes Timpani as an AI scheduling tool that builds nursing schedules automatically from algorithmic forecasts of patient volumes and staffing, and quotes HCA's innovation executive writing that its efficiency comes from AI generating stronger initial schedules, which distinguishes it from ordinary rota software by the operator's own account. The nurses say the tool assigned them shifts and days off against their requests, with the consequences they describe. HCA says nursing leaders make the final decisions, so the allocation is recorded both as the tool's decision (nurses' account) and as an AI output informing managers' decisions (HCA's account).","status":"reported"},"person_relations":["made_decision_about","informed_decision_about"]},"name":"US: HCA nurses tell WIRED its AI scheduling tool overrode their shift requests and scheduled them on requested days off; one delayed a medical appointment","summary":"Nurses at HCA Healthcare, the largest US hospital chain, told WIRED, in an article published on 2 October 2026, that Timpani, an AI scheduling tool HCA co-developed with Palantir and has rolled out at about 130 of its 190 locations, assigns shifts that ignore their requests. A Florida critical care nurse said that over four months the tool gave her different shifts for more than half of the 50 she requested, often back-to-back days that left her mentally drained. A Missouri nurse said essentially every nurse at their facility had been assigned to work on a requested day off, and that they delayed a long-planned doctor's appointment for a chronic condition after struggling to trade such a day. HCA says nursing leaders, not Timpani, make final scheduling decisions and that the tool schedules nurses on about 1 percent of requested days off.","incidentKind":"ongoing_experience","incidentDatePrecision":"unknown","exposurePattern":"repeated_interactions","reportedDate":"2026-10-02","aiSystem":"Timpani, an AI scheduling tool co-developed by HCA Healthcare and Palantir on Palantir Foundry, which uses algorithmic forecasts of patient volumes and staffing needs to build nursing schedules automatically","aiProduct":"Timpani","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["loss_of_autonomy","psychological_distress","treatment_disruption"],"harmOutcomeSummary":"According to the nurses interviewed by WIRED, the AI scheduling tool repeatedly assigned a Florida nurse shifts she had not requested, often back-to-back days that left her mentally drained, and assigned a Missouri nurse to work on a requested day off, leading Barker to delay a doctor's appointment for a chronic condition. HCA says managers make final decisions.","frameworkFacets":[],"causationStatus":"alleged","participantUsersAffectedMin":0,"otherPeopleHarmedMin":2,"affectedCountStatus":"partial","affectedCountEvidence":"Two counted: the Florida and Missouri nurses who describe specific consequences. Barker says essentially every nurse at the Missouri centre was assigned a requested day off, and five other nurses describe problems, but no further count with a stated consequence is given; patients are not counted.","victimAgeRange":"adult","jurisdiction":"US","platformType":"other","outcomeType":"media_coverage","outcomeStatus":"ongoing","primarySourceUrl":"https://www.wired.com/story/ai-making-mess-of-nurses-schedules-they-say-its-a-safety-issue/","primarySourceLabel":"WIRED, 2 October 2026: AI is making a mess of nurses' schedules","firstPublishedAt":"2026-10-13T03:27:43.266051+00:00","updatedAt":"2026-10-13T03:27:43.266051+00:00","scopeVersion":"facts-v3","tags":["workplace-ai","scheduling","nurses","hospital","palantir","union-grievance"]},{"id":"2026-cline-agent-cleanup-command-quoting-error-deleted-windows-data-drive-root-about-1-5-tb-first-person-issue","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'In Act mode, needed to clean a temp folder'; 'The nested escaping'; 'The path resolved to the drive root'; 'The text above was generated by the Cline AI agent on my behalf'","relation":"supports","source_id":"s1"}],"assertion":"The issue reports that, in Act mode, the Cline agent built a temp-folder cleanup command wrapping rmdir /s /q in cmd /c inside a PowerShell string, that the nested quoting was mangled and that the recursive delete ran against the G: data drive instead of the intended subfolder; the author says the agent wrote the issue text on their behalf.","causal_attribution":"The issue attributes the command to the agent's run_commands tool."},{"id":"c2","status":"reported","evidence":[{"locator":"'Result: recursive deletion of the drive's contents (~1.5 TB of the user's photos/videos/docs)'; 'only the code was recoverable (it was under git and pushed), user data was lost'; 'Recovery on an SSD was largely unsuccessful (TRIM)'","relation":"supports","source_id":"s1"}],"assertion":"The issue reports that about 1.5 TB of the user's photos, videos and documents and the project being edited were deleted, that only the code was recoverable from git, and that SSD recovery was largely unsuccessful.","causal_attribution":"User's account via the agent-written filing. The author later qualified the importance of the deleted data and reported a partial recovery (c5)."},{"id":"c3","status":"reported","evidence":[{"locator":"'I reproduced the mechanism in a sandbox. The path did not resolve to `G:`. The quoting gave `rmdir` a second target, `\\`, which is the root of the current drive, and `rmdir /s /q` deletes every target it is given.'","relation":"supports","source_id":"s1"}],"assertion":"A commenter reports reproducing the quoting mechanism in a sandbox: PowerShell ended the string at the first escaped quote, so rmdir received a bare backslash, the root of the current drive, as a second target.","causal_attribution":"Commenter's own sandbox test; establishes the mechanism, not the author's loss."},{"id":"c4","status":"reported","evidence":[{"locator":"'an Act-mode pre-tool guard for Windows PowerShell `run_commands`'; 'the existing command guard is registered only for Plan mode'","relation":"supports","source_id":"s1"}],"assertion":"A project contributor proposed an Act-mode pre-tool guard for Windows PowerShell run_commands and noted that the existing command guard is registered only for Plan mode.","causal_attribution":"Not applicable."},{"id":"c5","status":"reported","evidence":[{"locator":"'The PC was a new home machine, only a year old at the time of purchase, so I hadn't accumulated too many critical files in that year.'; 'a lot of that 1.5TB turned out to be junk anyway'; 'the mistake is entirely mine'; 'the model I was using was deepseek/deepseek-v4-flash'; 'P.S. I also managed to recover some files that were stored in other locations.'","relation":"supports","source_id":"s1"}],"assertion":"In a comment on 9 October 2026 the author wrote that the computer was a new home machine, only a year old, on which not many critical files had accumulated, that a lot of the 1.5 TB turned out to be junk, that some files stored in other locations had been recovered, and that the mistake was entirely the author's own; the author named the model as deepseek/deepseek-v4-flash.","causal_attribution":"The author's later statement; it qualifies the importance of the deleted material and reports a partial recovery, and does not withdraw the account of the deletion."},{"id":"c6","status":"reported","evidence":[{"locator":"'I don't think this will work.'; 'We decided it is better to simply have commands on or off, which is at least predictable.'; 'It looks like the model wrote `\\\"` when it should not have. So a stronger model or more prompting may help here.'; 'I'll bring it up with our product team.'","relation":"supports","source_id":"s1"}],"assertion":"A second contributor, writing for the project, replied that the proposed Act-mode guard would not work, that the project had decided to keep commands simply on or off, and that the model appeared to have written the escape sequence that split the command.","causal_attribution":"The contributor's reading of the commenter's reproduction; it attributes the faulty escape to the model and does not confirm the author's loss."}],"effects":[{"label":"about 1.5 TB of photos, videos and documents and the edited project deleted from the data drive; code recovered from git; the author later said much of the data was junk and some files were recovered from other locations (user's account)","claim_id":"c2","direction":"negative"}],"sources":[{"id":"s1","url":"https://github.com/cline/cline/issues/14864","kind":"first_person_account","access":"read","language":"en","translation_note":"Read in English on 2026-10-07: issue body and four comments retrieved through the GitHub API; re-read on 2026-10-11 with all 12 comments, including the second contributor's replies of 7 October and the author's comment of 9 October. The issue text is in English with a Cyrillic folder name and Russian-language Windows error strings quoted, which the research agent (an AI) read without translation. The author's handle is not recorded.","independence_group":"github-cline-14864-author"}],"version":1,"ai_roles":["own_use"],"contexts":["everyday_life"],"unknowns":["The date of the deletion; the issue gives none.","How much of the deleted personal data was permanently lost; the author later said a lot of it was junk and some files were recovered from other locations.","Where the author lives.","Whether Cline has shipped a fix."],"geography":{"basis":"The issue does not say where the author was or lives; a Cyrillic folder name and a Russian-language error message do not establish a country. No country is recorded.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Published under the charter's public first-person rule as a concrete account of an AI coding agent's own cleanup command deleting a user's personal drive contents, with the agent's authorship of the filing disclosed, the loss attributed to the user, the mechanism's reproduction attributed to a commenter and the project's reply to a contributor; uncorroborated as to the loss. The author's later qualification of how much of the deleted data mattered, the partial recovery and the author's acceptance of responsibility are recorded. The author's handle is not recorded.","reviewed_on":"2026-10-11"},"ai_involvement":{"basis":"The issue states that the Cline agent, in Act mode, constructed and ran the cleanup command whose mangled quoting sent rmdir /s /q against the drive root, and attributes the deletion of the drive's contents to that command; the author adopted the account by filing it and said the agent wrote the text on their behalf. A commenter's sandbox reproduction supports the quoting mechanism but not the author's loss. A second contributor, writing for the project, wrote: \"It looks like the model wrote `\\\"` when it should not have.\" In a later comment the author took responsibility (\"the mistake is entirely mine\") and named the model as deepseek/deepseek-v4-flash, without withdrawing the account of the agent's command.","status":"reported"},"person_relations":["acted_on_behalf"]},"name":"GitHub issue by a Cline user: a cleanup command the agent built ran against a Windows data drive's root, deleting about 1.5 TB of personal files, user says","summary":"In a GitHub issue filed on 6 October 2026 in the Cline repository, a user reports that a temp-folder cleanup command built by the Cline coding agent in Act mode on Windows was mangled by nested quoting and ran a recursive delete against the root of the G: data drive instead of the intended subfolder. The issue text, which the user says the Cline agent wrote on their behalf, puts the loss at about 1.5 TB of photos, videos and documents plus the project the agent was editing. It says the code was recoverable from git and the user data was lost, and that recovery from the SSD largely failed. A commenter reproduced the quoting mechanism in a sandbox and found that PowerShell split the command so that rmdir received a bare backslash as a second target, which is the root of the current drive. One contributor proposed an Act-mode guard. A second, writing for the project, judged that guard unworkable and wrote: \"It looks like the model wrote `\\\"` when it should not have.\" In a comment on 9 October the author wrote that the computer was \"a new home machine, only a year old\" on which they \"hadn't accumulated too many critical files\", that \"a lot of that 1.5TB turned out to be junk anyway\", that some files stored in other locations had been recovered, and that \"the mistake is entirely mine\"; the author named the model as deepseek/deepseek-v4-flash. The loss itself rests on the user's account.","incidentKind":"single_event","incidentDatePrecision":"unknown","exposurePattern":"single_interaction","reportedDate":"2026-10-06","aiSystem":"Cline coding agent, VS Code extension version 4.1.22, using its run_commands tool in Act mode to invoke cmd /c through PowerShell on Windows, per the issue; in a later comment the author named the model as deepseek/deepseek-v4-flash","aiProduct":"Cline","aiCompany":"Cline","severity":"low","verificationStatus":"unverified","harmCategories":[],"harmOutcomes":["property_loss","other_material_harm"],"harmOutcomeSummary":"The user reports that a recursive delete built by the Cline agent for a temp-folder cleanup ran against the root of their Windows data drive and deleted about 1.5 TB of photos, videos and documents plus the project being edited, with the code recoverable from git. The filing said the user data was lost. In a later comment the author wrote that a lot of the 1.5 TB turned out to be junk, that few critical files had accumulated on the machine and that some files stored elsewhere were recovered (agent-written filing adopted by the user and the user's later comment, uncorroborated).","frameworkFacets":[],"causationStatus":"alleged","participantUsersAffectedMin":1,"otherPeopleHarmedMin":0,"affectedCountStatus":"exact","affectedCountEvidence":"One person: the account's author. Exact 1.","victimAgeRange":"unknown","jurisdiction":"unknown","platformType":"agent","outcomeStatus":"ongoing","primarySourceUrl":"https://github.com/cline/cline/issues/14864","primarySourceLabel":"GitHub issue cline/cline #14864, 6 October 2026: mangled quoting led to rmdir /s /q running against a drive root, mass deletion of user data","firstPublishedAt":"2026-10-07T03:20:29.209252+00:00","updatedAt":"2026-10-11T07:43:49.254816+00:00","scopeVersion":"facts-v3","tags":["first-person","github-issue","cline","ai-agent","coding-agent","data-loss","file-deletion","windows","drive-root","powershell-quoting","acted-on-behalf"]}]}