AI News, 15 November 2026

Sunday 15 November

Google AI Edge Gallery adds offline speech

Google AI Edge Gallery adds offline speech

AI-generated summary Google introduced on-device automatic speech recognition and text-to-speech in AI Edge Gallery alongside reference code for local voice applications. Google reports an eight to nine percent word error rate on FLEURS for its system automatic speech recognition and about one hundred milliseconds before its system text-to-speech starts speaking.

Earlier:

How the Nvidia scandal impacts US–China relations: Department of Defense CTO

AI-generated summary US Secretary of Commerce Howard Lutnick will reportedly testify before the Senate Banking Committee on Dec. 10 amid questions regarding Nvidia and chip sales to China. US Department of Defense CTO Emil Michael discussed export controls and trade theories concerning chips during an interview with Deirdre Bosa.

Earlier:

Claude Agent Memory Explained: How Anthropic's "Dreaming" Fixes What Claude Remembers

Claude agent memory

AI-generated summary Anthropic published an engineering post on November 6, 2026, explaining how Claude keeps knowledge between sessions using a process called dreaming. The company organizes Claude agent memory into four layers, stacked from raw transcripts at the bottom to injected context at the top.

Earlier:

Canonical launches AI-generated security patch archive for Ubuntu Pro subscribers

AI-generated summary Canonical has introduced an opt-in software archive called Edge that provides Ubuntu Pro subscribers with early access to security patches generated by an AI multi-agent system named Surge. Kernel vulnerabilities are excluded from the Surge system, and the machine-written patches require manual activation rather than being enabled by default.

Muse Is Now in Chrome: Meta Quietly Publishes a Browser Extension

Jolly sitting on a sunlit windowsill inside a giant open browser window frame, with colourful web page cards and price tags floating around it, warm afternoon light, no text

AI-generated summary Meta has quietly published a Chrome extension called Muse from Meta on the Chrome Web Store. TestingCatalog reported the listing, which allows users to chat with Muse from any open page and have it browse using their personal Chrome.

Earlier:

Claude leads agent rankings as DeepSeek tops value scores

AI-generated summary Claude Fable 5.1 achieved the highest general-task Agent Fit rating of 79.6 in an AI Agent Store comparison reviewed on 2026-11-13. DeepSeek V4.1 Flash led the Practical Value category with a score of 84.7, pairing an Agent Fit rating of 78.9 with an estimated workload cost of $1.02.

Earlier:

LG Uplus unveils token‑optimization suite to cut AI agent costs for B2B customers

Source-provided image accompanying LG Uplus unveils token‑optimization suite to cut AI agent costs for B2B customers

AI-generated summary LG Uplus revealed a token-optimization suite designed to reduce operating costs for artificial intelligence agents used by business-to-business customers. LG Uplus CTO Lee Sang-yeop stated that the platform uses an in-house model, model routing, and a hardware strategy combining Nvidia, AMD, and FuriosaAI chips.

Korea's top tax exam scorer confesses he cheated using AI and a hidden camera

AI-generated summary Han Won-seok confessed to using generative AI and a hidden camera to achieve the top score on the national certified tax accountant exam in South Korea. The Seoul Metropolitan Police Agency received a written confession from the 34-year-old after he initially denied the allegations in an interview.

Also covered bySouth China Morning PostYonhap News AgencyNakoda News

Cairn AI Framework Hits Semiconductor and Defense Orgs

AI-generated summary Palo Alto Unit 42 reports that the Chinese-speaking threat cluster designated CL-STA-1238 used the open-source Cairn framework in twenty-five autonomous offensive cyber operations targeting semiconductor and defense organizations. The campaign relied on Zhipu GLM and DeepSeek models to drive agent workers through a shared blackboard architecture, successfully achieving unauthorized access or data retrieval at ten targets.