Image: embracethered.com
Worked Proof shown
Claude Pirate: data exfiltration through the Files API
A security write-up showing how network access in the code interpreter could leak data.
—
Earliest dated source:
The full story
Security researcher Johann Rehberger (Embrace The Red) showed that when Claude's code interpreter gained network access, an attacker could abuse it to steal data. Via indirect prompt injection, hidden instructions in content Claude reads could make it upload the user's data to the attacker's own account using Anthropic's built-in Files API. That route avoids the usual tricks with links. The post walks through the attack chain.
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Lemmy: Claude Pirate: Abusing Anthropic's File API For Data ExfiltrationPublished
The full write-up of the exfiltration chain.