-
Security incident 17 Nov 2026· 02:42 UTC Original report: 10 Nov 2026
Railway breach: stolen AI keys ran up $740,000; OpenRouter costs covered
Railway estimates $740,000 in unauthorized AI usage after customer API keys were stolen. It will cover OpenRouter-related losses after OpenRouter upheld its no-refund policy.
-
Security incident 16 Nov 2026· 22:53 UTC Original report: 10 Nov 2026
Microsoft: fewer than 10 edited neurons can hide a backdoor in AI agents
Microsoft’s Lazarus research shows that editing fewer than ten dormant neurons can hide a trigger-activated command payload in an LLM agent without retraining.
-
Security incident 16 Nov 2026· 16:33 UTC
Google: AI agents weaponize patches as “LLM Mixers” resell stolen API access
Google and Mandiant say attackers are using autonomous AI agents to accelerate intrusions and reselling hijacked LLM access through underground proxy services.
-
Security incident 16 Nov 2026· 15:36 UTC
Amazon: AI agents hunted LLM API keys in exposed gateways and apps
Amazon says AI agents targeted exposed LLM gateways and applications to harvest provider API keys, with stolen keys powering further credential hunting.
-
Security incident 16 Nov 2026· 13:30 UTC Original report: 12 Nov 2026
TypoPrompting campaign uses 186 fake AI domains to deliver malware and search hijackers
Island researchers traced 186 lookalike domains across 27 AI brands to ClickFix lures, browser-hijacking malware and search-hijacking extensions.
-
Security incident 12 Nov 2026· 02:37 UTC Original report: 11 Nov 2026
Notion says Fireworks AI breach exposed some Notion AI chat transcripts
Notion confirms that the Fireworks AI breach exposed AI chat transcripts from fewer than 1% of its workspaces, despite a contractual zero-data-retention commitment.
-
Security incident 10 Nov 2026· 15:32 UTC
NEAR AI: SGLang flaw exposed data of about 100 users, confidential queries may have been visible
An attacker used an SGLang vulnerability on Nov. 8 to reach NEAR AI infrastructure. TEE isolation did not stop a software bug inside the enclave.
-
Security incident 10 Nov 2026· 14:37 UTC Original report: 9 Nov 2026
Fireworks AI discloses second breach: billing data, datasets and incomplete model weights accessed
The LLM inference provider says attackers reached development machines via a hacked cloud provider on 24 October. Inference traffic is not reported affected so far.
-
Security incident 9 Nov 2026· 21:22 UTC Original report: 3 Nov 2026
OpenAI staff warned for years about internal security gaps, report says
Former employees describe broad code access and inconsistent reviews of internal tools. OpenAI says sensitive work is now segregated and tooling reviews are routine.
-
Security incident 9 Nov 2026· 20:44 UTC
South Korea reviews Anthropic breach affecting about 100 corporate Claude accounts
South Korea’s privacy regulator is reviewing Anthropic’s report of a breach affecting about 100 Korean corporate Claude accounts. Account and contact details may have been exposed; Anthropic says prompts, AI outputs and payment data were not accessed.
-
Security incident 9 Nov 2026· 17:03 UTC Original report: 5 Nov 2026
Claude Code permission files exposed passwords and API keys in public GitHub repositories
Apiiro found 1,122 public Claude Code settings files on GitHub containing credentials. Approved commands can preserve passwords and API keys in plain text; affected users should rotate secrets before deleting files.
-
Security incident 9 Nov 2026· 04:56 UTC Original report: 4 Nov 2026
Claude Dispatch flaw lets a browser extension run commands on the PC
Forever Security showed that an extension with access to claude.ai can drive Claude Desktop through Dispatch and run code on the machine. Anthropic called the report valid but a duplicate and is retiring Dispatch.