AI News

Meituan Open-Sources LongCat-2.5: 1.6 Trillion Parameters, Trained Entirely on Ascend, Surpassing GPT-5.6 on 20 Benchmarks

Meituan released the weights of LongCat-2.5 on November 15, 2026, a 1.6-trillion-parameter MoE model trained entirely on Huawei Ascend chips without touching Nvidia GPUs, outperforming GPT-5.6-Sol on 20 of 27 tests. The report examines the evaluation data, analyzes how domestic chips affect the ceiling of model capability, and discusses the industrial significance under export controls.

美团 LongCat-2.5 昇腾芯片
83

Cursor Key Leak Leads to Malicious Claude Code Extension Listing, Affecting 5,000 Developers

Cursor confirmed that a stolen access key issued to Fireworks AI enabled malicious Claude Code extension versions 2.1.312 and 2.1.313 to be listed, potentially affecting about 5,000 developers. The extension attempted to install persistent SSH backdoors, prompting Cursor to revoke credentials, rebuild its marketplace, and move all inference traffic away from Fireworks.

AI Safety 供应链攻击 Cursor
106

AI Security Arms Race: Anthropic Hires 33 Former Spies This Year; OpenAI Engineer Accused of Planting a Listening Device in Alleged Saudi Connection

Anthropic has hired 33 people with intelligence backgrounds this year as AI companies build counterintelligence teams, while a former OpenAI engineer faces charges of planting a covert recorder at its headquarters and allegedly contacting a Saudi royal. The cases show that LLM weights and strategic information have become state-level assets, pushing security from cyberspace into the physical layer.

AI Safety 工业间谍 Anthropic
126