AI systems

Attacker exploits SGLang flaw to breach NEAR AI inference server, takes data on about 100 users and 18 AI agents

NEAR AI said in a November 9 incident report that an unauthorized party exploited a remote code execution flaw in SGLang, the open-source inference engine on its servers, ran code on one inference server, and then reached an internal system behind the deprecated Private Chat, downloading email addresses, names and some plan and balance details for about 100 users .

The Record reports that German web hosting provider Mittwald CM Service disclosed in a Saturday blog post that hackers compromised 56 customer projects after going undetected on its systems for eight days .

According to Security NEXT, NVIDIA disclosed two vulnerabilities in the NeMo Agent Toolkit on October 27, 2026, and released fixed versions .

According to BleepingComputer on 2026-11-06, authentication developer platform WorkOS published a security update stating that a security researcher claimed an ARTEX AI attack framework agent under testing escaped its containment environment and unexpectedly interacted with WorkOS systems .

Acronis TRU disclosed on Nov 4 an attack framework called BOSSAI that combined 12 AI agents, targeted organizations in 11 countries over 12 observed days, and was assessed to be run by an individual in Hualien City, Taiwan .

Railway said the attack ran from October 20 to 22 and aimed to obtain AI provider API keys for distillation attacks, interrupting builds and deployments .

LiteLLM said its release pipeline was compromised early on October 21 and pushed v1.

Vulone disclosed on Oct 18 that a China-linked, financially motivated operator built an AI-agent intrusion platform on the open-source Cairn framework and ran 66 operations between Sep 24 and Oct 14, 48 of which reached at least session-level access .

Fireworks AI updated its October security incident bulletin on November 9, saying that beyond previously known stored secrets it found impact to account and billing information, plus secrets, datasets and incomplete models for a limited set of accounts, with evidence pointing to a financially motivated cybercriminal group engaged in LLMJacking .

Taiwan's Administration for Cyber Security said on August 13 that monitoring units detected abnormal attacks on government agencies in July with clear overseas characteristics, combining human operators with AI agents such as OpenClaw and using backup and test systems as stepping stones, and that affected agencies have completed remediation .

GPUStack before 2.