--- id: "2026-11-09-fireworks-ai-breach-datasets-weights" url: "https://postcutoff.com/e/2026-11-09-fireworks-ai-breach-datasets-weights/" as_of: "2026-11-22T13:32:00+01:00" date: "2026-11-09" date_precision: day category: policy-safety importance: 3 confidence: medium status: [Partly confirmed] sources: 4 editor: human editor human_review: null version: "2026-11-15" --- As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-11-09-fireworks-ai-breach-datasets-weights/ # Fireworks AI says hackers reached some customers' datasets, incomplete model weights and billing data AI inference provider Fireworks AI updated its breach disclosure on 9 November 2026: after intrusions detected on 6 and 24 October (the second via a security incident at an external cloud host), attackers reached billing and account data for customers and, for a small subset, datasets, confidential secrets and incomplete model weights. Fireworks links the activity to "LLMJacking". ## Key facts - Oct 6: intrusion detected, customers notified Oct 7; environment variables of evaluation jobs read (API keys, tokens) - Oct 24: second intrusion through an external cloud host into development machines, running an unauthorized workload - Exposed: billing and shipping addresses, invoices, payment history, balances, credit usage (no card numbers); for a small subset, datasets and incomplete model weights - Fireworks has not said how many customers were affected; earlier it had said it found no evidence of access to customer models or datasets - Cursor (BleepingComputer, Nov 14): a credential exposed in the Fireworks breach let an attacker push two trojanised copies of Anthropic's Claude Code extension (2.1.312 and 2.1.313) to Cursor's marketplace cache on Oct 29 for about 11 hours, reaching up to ~5,000 developers; under 0.05% of users had coding-agent traces exposed and under 0.01% API credentials; Anthropic said it 'didn't build or publish' that copy; Cursor rotated credentials and moved inference off Fireworks - TechCrunch (Nov 12): the Fireworks attack also let hackers access Notion customers' chat transcripts ## What happened Written from secondary security press; we did not read Fireworks' own bulletin. ## Why it matters Inference and fine-tuning platforms hold customers' training data and custom weights; this is one of the first disclosed breaches reaching model weights at such a provider. ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 162 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 132 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 223 days after its cutoff - Grok 4.7 (training cutoff May 2026): 162 days after its cutoff ## Sources 1. [Forbes: Fireworks cyberattack exposes customer data (headline)](https://www.forbes.com/sites/annatong/2026/11/09/fireworks-cyberattack-exposes) (forbes.com, press) 2. [Cybersecurity News: Fireworks AI breach via third-party cloud provider](https://cybersecuritynews.com/fireworks-ai-data-breach/) (cybersecuritynews.com, press) 3. [BleepingComputer: Cursor confirms security breach led to malicious Claude Code extension](https://www.bleepingcomputer.com/news/security/cursor-confirms-security-breach-led-to-malicious-claude-code-extension) (bleepingcomputer.com, press) 4. [TechCrunch: Cyberattack on Fireworks AI allowed hackers to access Notion customers' chat transcripts (Nov 12)](https://techcrunch.com/2026/11/12/cyberattack-on-fireworks-ai-allowed-hackers-to-access-notion-customers-chat-transcripts/) (techcrunch.com, press) ## Changes - 2026-11-15: Cursor/Claude Code extension and Notion follow-ups (news agent, 2026-11-14) - 2026-11-10 (filed): Created