hotai.news
AI Briefings / Date Archive

2026-11-14
AI Briefings Daily Report.

An archive of 83 AI news briefs collected on 2026-11-14. Check the original sources for details and updates.

This content still needs to be cross-checked against the original publication page.

83 curated itemsView other dates ↗
04News

Sierra launches Personal Agent Protocol v1.0; Manus notes its role in building agent ecosystem infrastructure.

Manus reposted Sierra's announcement: Personal Agent Protocol (Poppy) v1.0 launched on the day, available to all Meta AI agent users and every Shopify merchant, supported by over 100 design partners. Manus commented that the future of AI is not just smarter agents, but an entire action ecosystem built for agents, expressing pride in helping build the infrastructure that makes it possible.

Source links can be verified on the original site
07News

Google's Gemini personal AI features are now integrated with the Google Wallet app, providing personalized analysis and suggestions on user spending habits.

After user authorization, users can tag @Google Wallet in Gemini chats to inquire about financial status, spending habits, and budget planning. Gemini can also link credit cards and bank accounts via Plaid, generate charts for cash flow and asset allocation, and help create savings plans. However, it can only read information and cannot transfer funds or make investments.

Source links can be verified on the original site
12News

Microsoft veteran with 14 years of AI experience leaves: claims innovation struggles to be implemented and cost-reduction research is not taken seriously

Tech media outlet Windows Latest published an article today (November 14), reporting that John Langford, an AI scientist who worked at Microsoft for over 14 years and co-founded the AI Frontiers lab, has resigned. He stated that the key reason for his departure is that research results lack a clear path to practical application, and innovative ideas are often not taken seriously.

Source links can be verified on the original site
17News

Natura ring turns your fingers into an AI agent gateway

AI agents are right at your fingertips, from Claude to ChatGPT—and it's real. Natura's Interface is a ring that lets you trigger voice requests with a thumb press. NatureOS routes them to the AI agents you already use: Claude, ChatGPT, Grok Bot, Meta's Muse, or Instinct. You can choose which AI handles which task. Claude writes code; Instinct books restaurants. Responses return to your headphones or iPhone lock screen. No need to unlock your phone or open an app.

Source links can be verified on the original site
19News

The Tongyi app can now connect to continuous glucose monitors (CGMs) to analyze blood sugar fluctuations combined with daily records.

The Tongyi app now supports connecting to continuous glucose monitors. Users can complete authorization via 'Health Profile' → 'Bind Smart Device'. The app combines glucose data with daily records to analyze the causes of fluctuations. On World Diabetes Day, Tongyi demonstrated interpretation examples for scenarios such as temporary blood sugar spikes after exercise, rapid blood sugar rises from high GI foods like white rice and steamed buns, and the impact of sleep deprivation and stress on blood sugar regulation.

Source links can be verified on the original site
20News

Claude Code for Desktop now supports custom themes.

![Image 1: @ClaudeDevs](https://x.com/ClaudeDevs) ClaudeDevs ![Image 2: Anthropic](https://x.com/AnthropicAI) @ClaudeDevs 2h You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!

Source links can be verified on the original site
25News

Can gradient-aligned rewards make supervised learning policy-based? Baseten tests this with Qwen3-8B.

Baseten Base Labs explores whether gradient-aligned rewards can perform gradient updates on student self-sampling while retaining expert demonstrations as learning signals. The method scores the cosine similarity between the student rollout and the log-likelihood gradient of the expert demonstration, offering an efficient implementation requiring only two standard backpropagation passes. Experiments were conducted using Qwen3-8B on a four-digit Countdown task, with expert rollouts generated by Opus 5.5.

Source links can be verified on the original site
26News

Federal Reserve Financial Stability Report: AI Emerges as Wall Street's Top Financial Risk

The Federal Reserve released a financial stability report. A survey by the New York Fed found that 87% of market participants now list AI as their top financial risk, up from 50% in spring. The report notes that enthusiasm for superintelligence has pushed stock valuations higher. As of October, AI infrastructure debt exceeded $680 billion—five times the annual total for 2023 and accounting for roughly 31% of investment-grade corporate borrowing. Major banks face exposure close to $1 trillion in large technology firms, data center construction, and related financing instruments.

Source links can be verified on the original site
28News

Figma Weave on AI's role in creative workflows: What to automate and what to keep

Figma explores the boundaries of AI in creative processes, proposing that only tedious tasks should be handed over to AI while human judgment remains for thinking-intensive work. The article uses Figma Weave's creation of a Gaussian splat campaign with the video model Seedance 2.5 as an example, noting that generation is merely raw material requiring human refinement. Weave allows users to explore ideas on the canvas and convert them into reusable AI workflows.

Source links can be verified on the original site
30News

Anthropic updates Claude Code Projects: the coordinator is now more proactive and reports thread status.

Anthropic's ClaudeDevs account announced a major update to Claude Code Projects: Claude instances within projects will coordinate tasks across parallel threads. The new coordinator is more proactive, answering questions in project chats and updating the status of ongoing threads. Anthropic developer relations staff member Ado Kukic shared and praised this proactive update, calling it his go-to daily tool for using Claude.

Source links can be verified on the original site
31News

Grok Bot can now join meetings.

After an update, Grok Bot can now join meetings. Users can ask it questions, assign it tasks, or have it record team decisions. The author cited this update and joked about asking the bot to complete tasks it had been told to do during the meeting.

Source links can be verified on the original site
32News

To let general AI models like GPT-6 Astra serve as the 'brain' for robots: OpenAI announces a restructuring of its embodied intelligence team.

Fortune published an article yesterday (November 13) revealing that OpenAI has disbanded its independent robotics team and moved personnel to other departments within the company. OpenAI confirmed this organizational adjustment to the media but emphasized that it does not mean exiting the robotics field; robotics remains central to its long-term vision, with the restructuring aimed at accelerating R&D.

Source links can be verified on the original site
33News

Claude Code Projects gets a major update

We've made a major update to Projects in Claude Code! In projects, Claude coordinates your tasks across parallel threads. The coordinator is now more proactive. It answers questions in the project chat and notifies you of ongoing thread status.

Source links can be verified on the original site
34News

You can set reminders for any thread or let Claude remind you directly. Previously, the thread would disappear from your view. When time is up, it reappears and Claude brings it up in the project chat.

We've made a major update to Projects in Claude Code! In projects, Claude coordinates your tasks across parallel threads. The coordinator is now more proactive. It answers questions in the project chat and notifies you of ongoing thread status.

Source links can be verified on the original site
36News

Claude Code projects now support /btw and /fork

/btw and /fork are now available in projects. If you want to discuss something else while working, /btw creates a separate thread so your current one stays on topic. To try a different approach, /fork starts a new thread from your existing one.

Source links can be verified on the original site
37News

FAA expands AI air traffic control system SMART across the entire US

The FAA is expanding its $8.75 billion SMART system to cover airspace at all 22 national air traffic control centers in the US. The system predicts traffic flow and provides rerouting and sequencing recommendations. Since launching in the Washington DC area on September 21, it has rerouted 180 flights, saving an average of 10 minutes per flight and nearly 80 gallons of fuel per flight. A union representing over 15,000 controllers stated they were not involved in the system's design, testing, or deployment.

Source links can be verified on the original site
39News

Speculative questioning in the Jev pattern: merging follow-up questions into a single API call

Author shares best practices from allietheicon on speculative questioning: batching multiple questions into one request is often faster and cheaper, even when unsure if answers are needed. Benefits include completing the entire task in a single call, paying only once for tokens in context to avoid redundant state costs, and maintaining accuracy since questions within the same call do not interfere with each other's outputs; the diagram suggests adding follow-ups alongside the initial gate request, with code checking the gate first, using only necessary answers while ignoring the rest.

Source links can be verified on the original site
42News

Claude Code v2.1.325 released: fixes zstd decoding security vulnerability and multiple stability issues

Claude Code v2.1.325 addresses a security vulnerability in zstd-compressed response decoding and fixes multiple bugs, including incorrect sub-agent model switching, erroneous context length reporting by third-party gateways per turn, and lost recent conversation history upon session recovery. The update also improves organization policy interception prompts, resolves error messages for /security-review when used in bare repository worktrees, and corrects display and behavior issues with commands such as /stats, /insights, and /voice.

Source links can be verified on the original site
45News

DeepSeek is making its full-stack chat application capabilities available through an API for developers.

DeepSeek is opening its full App capabilities via API, including search, file reading, link handling, content safety, and voice features. Developers can activate built-in functions with minimal parameters or call individual APIs for OCR, ASR, TTS, and more. An open-source frontend demo (supporting DeepSeek chat history import) is available on GitHub and api-docs.deepseek.com.

Source links can be verified on the original site
48News

Claude Duet sends invitations to all Pro, Max, and Ultra subscribers on the US waitlist

Anthropic's Claude Duet update: All US Pro, Max, and Ultra subscribers on the waitlist as of yesterday have received invitations. The waitlist entry is at https://claude.com/product/claude-duet. The preview includes quota resets: 3 for Pro, 1 for Max and Ultra; official plans will soon open to more countries outside the US and continue gathering feedback.

Source links can be verified on the original site
49News

Nvidia crisis and on-site observations from the Cerebral Valley AI Summit just before Anthropic's IPO

Newcomer reports from the Cerebral Valley AI Summit in San Francisco: Jensen Huang resigns following disclosure of a second extramarital affair; Reuters reported a third that same day. Attendees largely believe Nvidia will be unaffected in the short term but is under ecosystem pressure; Databricks CEO calls it a blow to the entire ecosystem.

Source links can be verified on the original site
52News

Grok Bot has added the ability to join meetings.

The author shared and thanked @baltaaazr and @lingxi for announcing that Grok Bot can now join meetings. Users can ask it questions, assign it tasks, or have it record team decisions.

Source links can be verified on the original site
53News

Grok Bot now supports joining meetings and executing tasks.

Grok Bot can now join meetings to answer questions, assign work, or record team decisions. Author Lauren Tan notes this makes the robot smarter at gathering context on its own, allowing it to attend and take notes even without Starlink; users can start by simply asking the bot directly.

Source links can be verified on the original site
54News

vLLM cross-instance image cache ID collision vulnerability and its fix

vLLM suffers from a cross-instance image cache collision issue when sharing external KV cache across multiple servers: when prefix cache and image processing cache are both disabled, each server counts image labels starting from 1, causing different images to collide on the same key in shared cache. The second server hits this collision and returns cached data for another image, appearing as a normal cache hit without errors. The fix involves having each server include its own random component in image labels so cross-server labels remain distinct; the relevant PR is #67461.

Source links can be verified on the original site
55News

vstack performs formal verification of vLLM as an agent, identifying 17 bugs

UCSD Hao AI Lab has formally verified vLLM's newly released vstack agent using formal verification and discovered 17 bugs (still increasing), ranging from silent output errors to server lockouts. The team has publicly shared five of these bugs along with their corresponding fix PRs, and provided links to the vstack plugin and bug tracking table.

Source links can be verified on the original site
56News

vLLM streaming session out-of-bounds write vulnerability in shared cache

vLLM has a silent out-of-bounds write issue: when streaming input, it only checks the length of each new chunk but fails to verify the total session length. This causes token data from excessively long sessions to overflow into shared prefix cache blocks, altering output for subsequent requests that reuse these blocks without triggering any errors. The issue can be fixed by restricting worker writes and discarding tokens that exceed limits. See PR vllm-project/vllm#67457.

Source links can be verified on the original site
57News

Bug fix for vLLM mixed-model prefix cache zero-block issue

When vLLM uses mixed models with fine-grained prefix matching enabled, it marks cache blocks as reusable before they are fully written, causing the second request in the same step to prematurely copy from zero-filled blocks and continue from scratch. These erroneous blocks remain in cache for reuse by subsequent requests while responses appear normal. On GB200, pre-fix trials showed 32 instances of copying entirely zero sources; post-fix, this dropped to zero. The small fix delays requests with unwritten copy sources to the next scheduling step; see vllm-project/vllm#67451 for details.

Source links can be verified on the original site
58News

Grok Bot 1.0 launches on Super and other plans

Now available in Super, Ultra, all Team, and Enterprise plans, plus corresponding legacy plans. View all other new content for version 1.0: https://x.ai/news/grok-bot-1-0

Source links can be verified on the original site
59News

Anthropic has proposed ten guidelines for safety evaluation.

Safety evaluation is essential for understanding whether a model is ready for deployment, what protective measures need to be built around it, and where training and research efforts should focus. In our engineering blog, we have outlined ten recommendations on how to design, conduct, and report safety evaluations. https://www.anthropic.com/research/building-strong-safety-evaluations

Source links can be verified on the original site
60News

Gary Marcus: The generative AI bubble is bursting

Gary Marcus reviews his assessment since August 2023, concluding that generative AI lacks technological moats and profitability as the bubble bursts: Nvidia has fallen 29% over six months, Oracle down 35%, CoreWeave down 54%, and Anthropic's expected IPO valuation dropped from $2T to approximately $1.5T.

Source links can be verified on the original site
61News

OpenAI launches ChatGPT Wallet to enable AI agents to make payments on behalf of users.

OpenAI has launched ChatGPT Wallet, allowing AI agents to pay for users. The feature was previously abandoned due to limited flexibility after the Instant Checkout project. It is currently available only in the US via Dots, requiring a ChatGPT Pro subscription and an eligible US bank card; neither merchants nor agents can see the actual card number. Payments are processed using tokens generated by American Express, Mastercard, and Visa.

Source links can be verified on the original site
63News

Claude Code Stop hook tip: Use exit code 2 to trigger automatic retries when checks fail.

Lydia Hallie shares a Claude Code tip: Use a Stop hook to enable automatic retries after check failures. The key is ensuring the hook exits with code 2. Exit code 2 sends the hook's stderr back to Claude to trigger a retry, while exit code 1 is non-blocking and causes Claude to end immediately even if the check fails.

Source links can be verified on the original site
65News

Dwarkesh Patel: After the math avalanche, automated AI research is on the horizon.

Dwarkesh Patel updates his timeline on superintelligence. He previously thought it would take at least a decade, but now estimates the gap between current AI and the strongest human AI researcher is only two to three years. His median estimate places this milestone in 2028, with an earliest possible date of 2027.

Source links can be verified on the original site
69News

According to Bloomberg, several banks refused subscription orders from mainland Chinese and Hong Kong investors, either deeming their allocation chances low or invoking the U.S. export control rule ITAR. Anthropic declined to comment. The IPO is led by Goldman Sachs and Morgan Stanley, with an additional 20 banks participating; shares are expected to be priced on November 18 and then traded on Nasdaq under the symbol ANT.

According to Bloomberg, several banks refused subscription orders from mainland Chinese and Hong Kong investors, either deeming their allocation chances low or invoking the U.S. export control rule ITAR. Anthropic declined to comment. The IPO is led by Goldman Sachs and Morgan Stanley, with an additional 20 banks participating; shares are expected to be priced on November 18 and then traded on Nasdaq under the symbol ANT.

Source links can be verified on the original site
70News

OpenAI partners with American Express, Mastercard, and Visa to launch ChatGPT Wallet

OpenAI has launched the ChatGPT Wallet in partnership with American Express, Mastercard, and Visa. This allows dot to securely complete purchases on behalf of users. During payment, the agent cannot access real card details; every transaction undergoes a security check and requires user approval. The feature is now available to US Pro subscribers of dot starting today.

Source links can be verified on the original site
71News

Gemini's personal AI now integrates with Google Wallet

Gemini's new personal AI feature allows connection to Google Wallet to answer personal finance questions based on spending habits. Users can generate dynamic charts for cash flow, asset allocation, and budgets within the chat, identify forgotten subscriptions and recurring charges, and have Gemini flag price increases. They can also securely link credit cards and bank accounts via Plaid to set personalized suggestions like holiday budgets.

Source links can be verified on the original site
72News

Gemini Wallet connections can only read your account

@GeminiApp We built these tools with safety and privacy at their core. If you choose to connect your wallet, Gemini can only read your account; it cannot transfer, manage, or invest your funds. Learn more: https://goo.gle/3UBasU9

Source links can be verified on the original site
76News

OpenAI launches ChatGPT Wallet, allowing agents to make payments on behalf of users upon approval

OpenAI has launched ChatGPT Wallet, enabling agents (dot) to securely complete purchases for users with their consent. The system does not access actual bank card details; every transaction undergoes a security check. This feature is now available to dot agents for US Pro subscribers starting today. Ed Bayes, a product designer at Codex, shared the announcement and noted that the team has also rolled out several other experience enhancements. See https://openai.com/index/introducing-chatgpt-wallet/ for more details.

Source links can be verified on the original site
77News

This Week in AI: Who benefits from AI advancements?

OpenAI introduces the LifeVal benchmark to measure AI capabilities in daily tasks such as childcare, household budgeting, and government services. Scores for GPT-5.2 stand at 49%, while GPT-5.6 and GPT-6 Astra rise to approximately 80%. Anthropic is now testing read-only bank connections within Claude; some US subscribers can analyze spending. A METR report suggests AI assistance could double progress on certain model development tasks for Anthropic.

Source links can be verified on the original site
82News

Gemini 4 Carbon creates a 3D Chrome Dino game.

We built a full 3D canyon adventure for Chrome's offline Dino using Gemini 4 Carbon at @Antigravity 🦖 Starting with flat character art made by Nano Banana, Carbon rigged the 3D models in Blender and developed a playable third-person desert parkour game featuring slide-jump-kick mechanics and day-night sky cycles.

Source links can be verified on the original site