Hijacked by injectionPI-0038
Hidden text in a document made Anthropic's Claude Cowork upload user files to an attacker's account
Days after Cowork launched, PromptArmor showed that near-invisible instructions in a document could make the agent upload files from a folder the user had connected. The files went out through an Anthropic web address the agent was allowed to reach. Johann Rehberger had reported the same route for sending data out through the file-upload interface in Claude.ai to Anthropic in October 2025, and it was not fixed before Cowork launched, according to reports.
Harm: No harm reportedControl level 2, Instruction violation
Not counted · can set the control floorEffect on the index
Near miss. Harm was plausible but prevented. It is rated only on what actually happened.
Sources
How we know
5 sources · single source. Links go to the original publishers; the summary above is in our own words.
- researchClaude Cowork Exfiltrates FilesPromptArmor · Jan. 14, 2026promptarmor.com/resources/claude-cowork-exfiltrates-files
- researchClaude Pirate: Abusing Anthropic's File API For Data ExfiltrationEmbrace The Red (Johann Rehberger) · Oct. 28, 2025embracethered.com/blog/posts/2025/claude-abusing-network-access-and-anthropic-a…
- newsContagious Claude Code bug Anthropic ignored promptly spreads to CoworkThe Register · Jan. 15, 2026theregister.com/2026/01/15/anthropics_claude_bug_cowork/
- newsClaude Cowork hit with file-stealing prompt injection days after Anthropic's launchThe Decoder · Jan. 17, 2026the-decoder.com/claude-cowork-hit-with-file-stealing-prompt-injection-days-afte…
- blogClaude Cowork exfiltrates files (link post)Simon Willison's Weblog · Jan. 14, 2026simonwillison.net/2026/Jan/14/claude-cowork-exfiltrates-files
Why this rating
No harm reported; 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
No harm reportedResearcher demonstration on controlled accounts; no victims reported.
Not a finding that no harm occurred.
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 agent used network access and files it had been granted, on injected instructions.
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
No external harm was reported from this controlled test. Control remains level 2. No Anthropic confirmation of this test appeared in the sources reviewed.
Effect on the index
It does not count toward the Jan. 19 reading
The reading for the week to Jan. 19, 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.
Not counted: no harm reported. 3 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | No | — |
| Worst documented harm, ksets the band | — (none counting) | — (none counting) |
| Harms at that level, nposition in the band | 0 | 0 |
| Highest control level breachedsets the reading only when no harm counts | 4 | 4 |
| Readingrounded down | 4 Control failures only | 4 Control failures only |
Inside the window of 4 weekly readings
| Week to | Reading | Band |
|---|---|---|
| Jan. 19, 2026 | 4 | Control failures only |
| Jan. 26, 2026 | 4 | Control failures only |
| Feb. 2, 2026 | 2 | Control failures only |
| Feb. 9, 2026 | 6 | Negligible harm |
Revisions
What we changed
1 logged. Every change to a rating is logged here, with the reason.
- v1Oct. 7, 2026
First published.
Cite and share
Use this record
Citation
Paperclip Index. “Hidden text in a document made Anthropic's Claude Cowork upload user files to an attacker's account.” Record PI-0038. Reported Jan. 14, 2026; updated Oct. 7, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0038