NEAR AI discloses SGLang vulnerability exploit, touts TEE as the fix
AccBalanced · x · 2026-11-11
NEAR AI revealed that on November 8 an attacker exploited a vulnerability in SGLang, its open-source inference engine, to access part of its infrastructure. Access has been closed, affected users contacted; NEAR Protocol and Intents were unaffected.
Founder ilblackdragon argues inference providers are now a honey pot for attackers — holding valuable data and the ability to inject actions into user devices via agents. He advocates TEE architecture so even the provider can't access user data, with open code letting partners verify what runs inside the enclave.
More from Infra
- Running a local coding agent: Qwen3.8-27B + DFlash2 with KV cache reuse on Apple Silicon — fuzhongkai · 2026-11-11
- HBM4 doubled pins to 2048; HBM5 will chase speed per pin instead — AccBalanced · 2026-11-11
- Anthropic says Claude ported itself to AMD GPUs in 3 days, writing ~100% of kernels unassisted — AccBalanced · 2026-11-11
- Polymarket puts 17% odds on orbital AI data centers by 2027 as Google launches Suncatcher — Polymarket · 2026-11-11
- Developer drops $10B Michigan data center after residents reject $54M tax promise — Polymarket · 2026-11-11
- Rumor: DeepSeek to pretrain ~2T-param V5 on Huawei Ascend 950, release may slip to 2027 — teortaxesTex · 2026-11-11