AI News Nov 11, 2026
Language: DE / EN / ZH / FR / ES / PT / JA / KO
AI-generated: summaries written by AI from the linked sources. How we use AI
Key Takeaways
- Google launched the Gemini Decisions API, powered by its new Gemini Decision 1 model, which returns predefined answers with confidence scores and can evaluate up to 1,000 yes/no or…
- AWS released AWS Emulator, a free tool that simulates more than 200 of its services locally. AWS says AI agents wrote the tool and that its team has not written or reviewed any of …
- Anthropic says it used its in-house Mythos model to port Claude to AMD GPUs in three days with almost no human intervention, a switch it says would normally take months. Anthropic …
- Applied Intuition: $1.5B (Unknown)
- When using a decision-style model that returns probabilities, pass multiple yes/no questions together with the media file and read the probability scores instea…
Why Today Matters
- AI accelerating engineering and verificationAnthropic says its Mythos model ported Claude to AMD GPUs in three days, while Google DeepMind says it mathematically verified parts of Signal's encryption in four weeks. Both claims, if accurate, point to AI compressing work that normally takes months.
- Agent security and policy gapsUnited Airlines is banning AI agents from booking flights, citing security and federal-rule risks, as Grok Build reportedly leaks data from over 35,000 apps via shareable links. Agent adoption is outpacing the guardrails around it.
- AI-written infrastructureAWS says AI agents wrote its free Emulator tool covering 200-plus services without human code review, while Anthropic reports near-unattended porting work. Vendors are now marketing limited human oversight as a feature rather than a caveat.
- Training data export loopholesChinaTalk reports US data brokers including Mercor and Surge AI are selling roughly $500 million in AI training data, including RL environments built for OpenAI or Anthropic, to Chinese labs. Export controls barely apply to this trade.
- China on-device and model efficiencyHuawei detailed a 30-billion-parameter mixture-of-experts model running offline on its Kirin 9050 phone, while Baidu previewed ERNIE 5.5 Flash with a 1M-token context. Chinese firms are pushing efficiency and local inference as competitive ground.
What are the top AI breakthroughs?
This Nov 11, 2026 covers 10 curated AI news items spanning technology, research, and product developments. Google launched the Gemini Decisions API, powered by its new Gemini Decision 1 model, which returns predefined answers w...
Google launched the Gemini Decisions API, powered by its new Gemini Decision 1 model, which returns …
Google launched the Gemini Decisions API, powered by its new Gemini Decision 1 model, which returns predefined answers with confidence scores and can evaluate up to 1,000 yes/no or choice options. Pricing is set at $0.10 per 1 million input tokens, positioning it for high-volume classification and decision tasks.
AWS released AWS Emulator, a free tool that simulates more than 200 of its services locally. AWS say…
AWS released AWS Emulator, a free tool that simulates more than 200 of its services locally. AWS says AI agents wrote the tool and that its team has not written or reviewed any of the code, which the company is touting as a selling point.
Anthropic says it used its in-house Mythos model to port Claude to AMD GPUs in three days with almos…
Anthropic says it used its in-house Mythos model to port Claude to AMD GPUs in three days with almost no human intervention, a switch it says would normally take months. Anthropic now runs Claude on four chip families and aims to cut costs and reduce vendor dependencies, claims that, if accurate, would erode Nvidia's CUDA moat.
Google DeepMind says it mathematically verified key parts of Signal's encryption in a four-week proj…
Google DeepMind says it mathematically verified key parts of Signal's encryption in a four-week project, a result that hints AI could make formal verification far more practical. The work was conducted with the Signal protocol as its target.
United Airlines said it is banning AI agents from making flight reservations on behalf of travelers,…
United Airlines said it is banning AI agents from making flight reservations on behalf of travelers, citing security risks and the possibility that such actions could violate federal rules. The policy arrives as consumers increasingly use agents like Meta's Muse and OpenAI's dots to complete tasks such as purchases and bookings.
Researchers found that Grok Build, the vibe-coding agent from SpaceXSI, is leaking data from over 35…
Researchers found that Grok Build, the vibe-coding agent from SpaceXSI, is leaking data from over 35,000 apps via shareable links, including proprietary business information and other sensitive secrets, some belonging to SpaceXSI itself. The finding was reported by Ars Technica.
Helsing unveiled Clarity, an AI platform that uses agents to distill sensor data into target recomme…
Helsing unveiled Clarity, an AI platform that uses agents to distill sensor data into target recommendations for weapons systems, under human oversight and European control. The company says every decision will carry a human name, though reporting notes that automation speed and data fusion create risks that are hard to contain.
According to a ChinaTalk investigation, US data brokers including Mercor, Surge AI, and AfterQuery a…
According to a ChinaTalk investigation, US data brokers including Mercor, Surge AI, and AfterQuery are selling roughly half a billion dollars' worth of AI training data, including RL environments previously built for OpenAI or Anthropic, to Chinese AI labs. At least nine firms are involved, and export controls barely apply.
Baidu released ERNIE 5.5 Flash Preview on its Qianfan API, a 280B-total, 14B-active model with a 1M-…
Baidu released ERNIE 5.5 Flash Preview on its Qianfan API, a 280B-total, 14B-active model with a 1M-token context window. Baidu says it will publish open weights by November 25, though a public download could not be confirmed.
Huawei detailed how its Kirin 9050 chip runs a 30-billion-parameter mixture-of-experts model on a ph…
Huawei detailed how its Kirin 9050 chip runs a 30-billion-parameter mixture-of-experts model on a phone without an internet connection, activating about 2 billion parameters at a time. The chip's Da Vinci NPU uses low-bit quantization and coordinates memory and compute to run the model locally.
Top AI Videos This Week
2 curated YouTube videos about AI developments.
Based on a transcript excerpt: Matt Wolfe interviews Peter Yang, a former product manager who quit h…

Based on a transcript excerpt: Matt Wolfe interviews Peter Yang, a former product manager who quit his job to become a full-time creator, about the personal AI system he uses to run his one-person business. Yang demonstrates how he connects tools like ChatGPT and Claude Code to create an AI 'chief of staff' that reads his email, calendar, and Slack twice a day, plus automated workflows for podcast research, video editing, and content creation. Wolfe also builds simplified versions of these systems so viewers can follow along and set them up.
Based on a transcript excerpt: AI Explained analyzes the Fable 5.2 and Gemini Carbon model releases,…

Based on a transcript excerpt: AI Explained analyzes the Fable 5.2 and Gemini Carbon model releases, drawing on a 264-page system card, a 30-page METR report on AI acceleration, and the creator's own testing. The video covers benchmark results, a 'Max Bench' test designed to catch models that memorized answers, the remote labor index, and claims from Anthropic that the pace of AI improvement will quicken. It also discusses whether benchmark results hold up to scrutiny.
What are the latest AI investment signals?
Latest AI investment signals: 5 funding rounds, 2 market updates, and 0 M&A transactions.
Primary Market – Funding Rounds
| Company | Amount | Round | Investors |
|---|---|---|---|
| Applied Intuition | $1.5B | Unknown | General Catalyst |
| Fractile | $720M | Unknown | Thrive, Founders Fund |
| Torus | $500M | Series C | Magnetar |
| Edelbyte | $70M | Series A and Seed | |
| Rox | $100M | Series B | General Catalyst, March |
Secondary Market – Market Updates
M&A – Mergers & Acquisitions
No M&A data.
What are practical AI tips this week?
5 practical AI tips curated from Reddit communities and expert blogs. Simon Willison tested Gemini Decision 1 on a video by passing several yes/no questions (dragon, robo...
Prompt Tips
Simon Willison tested Gemini Decision 1 on a video by passing several yes/no questions (dragon, robot, France, USA) alongside the video file; the model returned probability values rather than strict 1.0/0.0 answers.
When using a decision-style model that returns probabilities, pass multiple yes/no questions together with the media file and read the probability scores instead of expecting binary answers.
Productivity
Willison used tree-sitter to analyze the code of his vibe-coded tools for JavaScript APIs, libraries, and web API hosts, then added tag clouds to browse them by category.
Use tree-sitter to statically analyze a codebase for the APIs and libraries it uses, then generate tag-based browsing from the extracted tags.
Productivity
LLM 0.37.2 fixed a security issue where a configured OpenAI API key could accidentally be sent to third-party OpenAI-compatible endpoints, e.g. if the key named by api_key_name was never stored or a responses: true model was run with -o chat_completions 1.
When pointing a tool at third-party OpenAI-compatible endpoints, verify your API key is actually stored under the configured api_key_name and be careful with responses: true models run with -o chat_completions 1, to avoid leaking your OpenAI key.
Productivity
JioStar's Vijay Seshadri shared lessons on evals, guardrails, and taking AI into production at 500M-user scale, according to OpenAI News.
Before taking an AI feature to production, build evaluations and guardrails as part of the rollout process.
Productivity
Google DeepMind used AI to generate 6.5 million lines of mathematical proofs to check Signal's encryption code in four weeks, and restored two previously fixed security bugs to test whether the system could catch them, according to The Neuron.
To validate an AI-based bug-detection approach, reintroduce known previously fixed bugs and check whether the system catches them.