CVE-2026-82533: DeepSeek Harness Vulnerability Lets AI Agents Escape Their Own Sandbox
Original CVE-2026-82533 disclosure: agents could disable their own sandbox without approval.
4 items · 2 sources · 4 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
A third outlet (forkast.news, Sep 10) is still describing the same Sep 8 sandbox-escape flaw, not a new development or a fix.
Security researchers disclosed CVE-2026-82533 on September 8: a flaw in DeepSeek's open-source agent harness that lets an AI agent disable its own file-system sandbox without approval. Two more outlets repeated the same finding through September 10 with no vendor patch confirmed.
Arc
Original CVE-2026-82533 disclosure: agents could disable their own sandbox without approval.
A second outlet reframes the same flaw as a file-sandbox bypass.
A third outlet repeats the same disclosure two days later; still no patch mentioned.
Unrelated same-week HN post on building a custom agent harness -- grouped here only by shared wording, not the same event.
Original CVE-2026-82533 disclosure: agents could disable their own sandbox without approval.
A second outlet reframes the same flaw as a file-sandbox bypass.
A third outlet repeats the same disclosure two days later; still no patch mentioned.
Unrelated same-week HN post on building a custom agent harness -- grouped here only by shared wording, not the same event.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.