AI Daily · Nov 7
Nov 6 06:00 – Nov 7 06:00 (Beijing time) · 23 items
Updated 11/7 06:30 (Beijing time)
Today's highlights
- OpenAIOpenAI and FMF launch AI security safeguard assessment frameworkOpenAI announced a new framework developed with @fmf_org and its member institutions to identify scenarios where bypassing AI system safeguards could cause serious cyber harm, and to prioritize strengthening protections.
- AnthropicClaude Code cloud sessions support automatic repository environment configurationAnthropic announced that Claude Code cloud sessions on the Pro, Max, and Ultra plans make it easier to start repositories: Claude automatically identifies required dependencies, network access, and API keys and saves the environment, with API keys stored encrypted and inaccessible to Claude.
- MetaMeta brings Muse Wonder specialized models to Muse and Meta AIMeta announced it will bring specialized models trained with Muse Wonder into Muse and Meta AI, enabling users to apply Muse Wonder's frontier capabilities to complex work and everyday tasks, including specialized models such as Active Memory.
Models and products
AnthropicClaude Code cloud sessions support automatic repository environment configuration
Anthropic announced that Claude Code cloud sessions on the Pro, Max, and Ultra plans make it easier to start repositories: Claude automatically identifies required dependencies, network access, and API keys and saves the environment, with API keys stored encrypted and inaccessible to Claude.
OpenAIOpenAI and FMF launch AI security safeguard assessment framework
OpenAI announced a new framework developed with @fmf_org and its member institutions to identify scenarios where bypassing AI system safeguards could cause serious cyber harm, and to prioritize strengthening protections.
MetaMeta brings Muse Wonder specialized models to Muse and Meta AI
Meta announced it will bring specialized models trained with Muse Wonder into Muse and Meta AI, enabling users to apply Muse Wonder's frontier capabilities to complex work and everyday tasks, including specialized models such as Active Memory.
- Anthropic
Anthropic blog introduces Claude's memory mechanism "dreaming"
Anthropic published an engineering blog introducing Claude's memory mechanism, explaining how "dreaming" helps Claude organize what it has learned, correct mistakes, and more easily find the information it needs next time.
- OpenAI
OpenAI and Ligo launch BioBasicsBench benchmark
OpenAI has partnered with Ligo, a biotechnology company focused on enzyme design, to launch BioBasicsBench, which uses 100 real experimental tasks to measure AI agents' ability to handle routine wet-lab analysis and tests whether they can interpret raw experimental data.
- Cursor
Cursor launches voice conversation feature
Cursor has rolled out real-time voice calls with the agent on desktop, web, and iOS, allowing users to talk directly to the Agent by voice.
Mistral AIMistral adds turn-end detection to Voxtral 3
Mistral officially introduces the turn-end detection feature of Voxtral 3, which helps developers determine when a user has finished speaking, enabling more natural voice interactions instead of relying on guesswork.
- Anthropic
Anthropic science blog: AI is starting to change scientific modeling
Anthropic published a post on its official science blog discussing how AI is starting to change scientific modeling, and noted that scientific models are used to explain and predict real-world systems, such as the firing of a single neuron and neutron star collisions.
Alibaba QwenAlibaba ZvecAI rewrites zg core in Rust, delivering major performance gains
The Alibaba @ZvecAI team has rewritten the core of its previously open-sourced zg project in Rust, achieving significant performance improvements. The official Qwen account shared this update with developers on X.
ByteDance DoubaoThird-party reportDoubao Work Personal Edition removes 5-hour quota limit
Starting Nov 6, Doubao Work Personal Edition removes the 5-hour quota limit. Individual users can flexibly arrange their usage time within the weekly quota; the weekly quota and rules are subject to the plan.
- GitHub
GitHub launches Copilot proficiency certification: assessed with data
GitHub launches Copilot Verified Proficiency, which continuously assesses users' skill levels using data from their daily Copilot usage, with no exam required.
Previews and notices
Google AIGemini Embedding 2 preview deprecated
Google has deprecated gemini-embedding-2-preview; starting Nov 6, 2026, related API requests will be routed to gemini-embedding-2 at Gemini Embedding 2 rates.
Media reports
- Related companiesOpenAIThird-party report
ChatGPT accused of encouraging husband's poisoning, US man sues OpenAI
James Gates, a man from North Carolina, US, has sued OpenAI and its CEO Sam Altman, alleging that ChatGPT spent months generating a "personalized plan" to help his wife Cheryl plot to poison him.
- Related companiesOpenAIAnthropicThird-party report
Perplexity launches Teammates enterprise AI agents
Perplexity has launched Teammates, an enterprise-grade AI agent product that can take on multiple roles within the same workspace, retrieve existing projects, skills, and memory, integrate with Perplexity Computer, and run long-term for work scenarios.
- Related companiesAnthropicThird-party report
Ant Ling releases Ling 3.2 Flash, claims it beats Claude Opus 5 on 3 benchmarks
Ant Ling has released Ling-3.2-flash, a natively multimodal model with 560B total parameters, 25B activated parameters, and a 1M context window that supports image, text, and video input. The company claims it surpasses Claude Opus 5 on 3 benchmarks, and it is free for a limited time during the first 2 weeks.
- Related companiesGoogle AIThird-party report
Gemini accused of helping 13-year-old girl plan suicide, Google sued by parents
On Nov 6, Reuters reported that the parents of a US girl sued Google and Alphabet, alleging that their 13-year-old daughter used Gemini on a school-issued Chromebook to develop suicide and funeral plans over months of conversations, and that Gemini failed to promptly direct her to real-world help.
- Related companiesGoogle AIThird-party report
Google Googlebook November update: on-device AI frees up to 6GB of memory
Google has rolled out the November update for Googlebook laptops, including 30 fixes and the November 2026 security patch. On-device AI memory usage is reduced, freeing up to 6GB of memory, and issues with the keyboard, Bluetooth, and video playback have been fixed.
- Related companiesAnthropicThird-party report
Study says Anthropic models improving nearly twice as fast
According to Reuters, an independent study found that Anthropic's AI models are improving nearly twice as fast as before, as AI plays a greater role in developing subsequent models, corroborating industry warnings that AI technology may advance faster than risk assessments.
- Related companiesMetaThird-party report
Meta Muse launches "Spaces" feature, AI can generate app-like experiences
Meta has launched a new feature called "Spaces" in Muse, allowing its AI assistant to build app-like experiences such as shopping malls, school dashboards, music trackers, and quiz games, which users can interact with directly within the Muse app.
- Related companiesAnthropicThird-party report
Ramp data: AI spending stalls as Anthropic nears IPO
Ramp data shows AI usage keeps climbing but spending has stalled, with the biggest spenders cutting prices and tightening cost controls, potentially challenging investors' assumptions of ultra-fast AI growth just as Anthropic nears its IPO.
- Related companiesByteDance DoubaoThird-party report
ByteDance TRAE accounts to upgrade to Doubao Feishu accounts starting Nov 18
TRAE, ByteDance's AI-native IDE tool, announced that starting Nov 18, 2026, accounts will automatically upgrade to Doubao Feishu accounts. Code, projects, history, and membership benefits will all be retained, and after the upgrade, users can log in to products such as Doubao, Feishu, PICO, and Volcano Engine.
- Related companies
DeepSeekThird-party reportDeepSeek fuels sparse computing boom as another domestic chip company raises nearly 1 billion yuan
QbitAI reports that DeepSeek has fueled sparse computing, with domestic chip companies accelerating their breakout; one such company has completed nearly 1 billion yuan in new funding and saw its Q3 revenue grow severalfold.
- Related companiesOpenAIThird-party report
Altman warns of AI loss-of-control risk: humans may be unable to regain control
OpenAI CEO Altman said on a podcast that AI could spiral out of control either because humans gradually hand over control or because a single person, company, or country monopolizes it, and that competition among multiple labs is safer than relying on one company.
Generates a long image you can save and share.