Cursor breach traced to exposed Fireworks key: 0.01% credentials leaked, Claude Code extension swapped
AccBalanced · x · 2026-11-14
Cursor published a full post-mortem on its security incident. The attack began at partner Fireworks: an exposed access key let the attacker copy a read-only bucket of coding agent traces (mostly 2025, non-Privacy-Mode sessions). Some traces contained unredacted credentials, exposing API keys of fewer than 0.01% of customers. Using those, the attacker reached select SaaS tools, archived non-production source repos, limited support data and emails, usage metrics, and conversation summaries—and had earlier planted a malicious copy of Anthropic's Claude Code in the Cursor marketplace. Cursor says most customers were unaffected and has notified those impacted.
More from Safety
- Bloomington, Indiana Bans Superintelligent AI, Hoping 350 Other Cities Will Follow — kimmonismus · 2026-11-15
- HiveAttest adds pre-action attestation for AI agents with signed receipts per call — modelcontextprotocol · 2026-11-15
- Anthropic book settlement checks arrive — one author's payout covers a daily Venti latte till summer — dfinke · 2026-11-15
- A week of chaos: Microsoft outage, Notion and Cursor hit, unverified Nvidia CEO resignation claim — tech__unicorn · 2026-11-15
- Witten, Tong, Maldacena sign statement urging AI companies to openly release HEP research results — kylekabasares · 2026-11-15
- Geoffrey Hinton renews warning: superintelligent AI may be impossible to keep under control — VraserX · 2026-11-15