In brief
- AI is becoming more deeply embedded in concrete workflows: Google is developing a model for rapid decisions, while Databricks’ Genie is intended to take on multistep enterprise tasks.
- Compute remains a business model under pressure: OpenAI’s revenue is growing significantly, but CoreWeave’s high financing costs and new chip investments show how expensive infrastructure expansion can be.
- Security and control are coming to the fore: Data leaks from AI-generated apps and debates about military AI make robust access controls and genuine human oversight operational requirements.
- Robotics is opening up to new applications: Serve Robotics plans to open its delivery network to external developers; research into navigation and manipulation is also aiming to make systems more reliable in real-world environments.
🔬 Models & Research
Google DeepMind’s AlphaProof Code mathematically proves that Signal’s encryption is secure
11.11.2026
DeepMind used AlphaProof Code to check Signal’s cryptographic core code as actually executed against specified security requirements. The approach could bring formal verification closer to the code running on devices, rather than keeping it limited to theoretical protocol designs.
Google’s AlphaGenome 2 investigates the cellular origins of disease
11.11.2026
DeepMind introduces AlphaGenome 2, a model for analyzing how changes in DNA affect gene activity across different cell types. More precise predictions could help researchers better pinpoint disease mechanisms and identify potential treatment approaches.
💼 Business & Markets
British chip developer Fractile raises $720 million
11.11.2026
British AI chip developer Fractile has raised $720 million in a Series C round, with support from the UK’s Sovereign AI Fund. Alongside its own hardware, the company is developing a compiler and positioning itself against the high inference costs of existing platforms.
CoreWeave turns to prepayments and shorter contracts to finance AI growth
11.11.2026
CoreWeave is relying more heavily on prepayments, loans tied to customer contracts, and shorter terms to finance its expansion. With revenue growing rapidly but losses and interest costs remaining high, the viability of its AI cloud business increasingly depends on sustained demand and pricing power.
OpenAI raises quarterly revenue to $10 billion—but losses remain high
11.11.2026
OpenAI is said to have increased its quarterly revenue by around 50 percent to $10 billion, The Decoder reports, while continuing to incur substantial operating losses. The growth demonstrates strong demand, but profitability and infrastructure costs remain central questions as the company scales further.
AI cuts the switch to AMD chips to three days, according to Anthropic
11.11.2026
According to Anthropic, the company can now run its models on several major chip families and use compute resources largely interchangeably. AI-assisted porting of compute kernels to AMD hardware reportedly took just three days, potentially reducing dependence on Nvidia’s CUDA ecosystem.
US companies’ AI spending is growing more slowly as new competition intensifies pressure
11.11.2026
According to the Ramp AI Index, US companies’ AI spending rose by just 1.65 percent month over month in October. Spending remains concentrated among a small number of companies and software teams, putting greater pressure on providers to demonstrate the concrete productivity gains their products deliver.
Helsing’s AI platform Clarity aims to be Europe’s answer to Palantir’s Maven
11.11.2026
Helsing has unveiled Clarity, a platform that combines sensor data and reports to support military decision-making. The initiative brings together European sovereignty goals and an area of application where transparency, human approval, and the consequences of errors are especially critical.
Plurai wants to improve AI agents using methods from autonomous driving
11.11.2026
Startup Plurai aims to train AI agents using simulations and synthetic data tailored to a company’s digital systems. This would allow agents to test complex workflows more often before deployment, rather than revealing errors only in production environments.
⚖️ Policy & Regulation
US data brokers sell $500 million worth of AI training data to Chinese labs
11.11.2026
According to an investigation, at least nine US companies are selling $500 million worth of training data to Chinese AI labs. The largely unregulated trade raises questions about export controls and potential risks in security-sensitive research fields.
UK regulator threatens universities with sanctions over lack of AI-proof exams
11.11.2026
According to a report, England’s higher education regulator is calling on universities to adapt exam formats to the widespread use of generative AI. Universities will need to strengthen assessment by relying more on supervised exams and oral presentations.
Grok apps expose internal SpaceXSI and X data, as well as patient data
11.11.2026
According to security firm Claroc, nearly 95 percent of more than 37,000 web apps created with Grok could be accessed without logging in, including apps containing sensitive corporate and patient data. Insecure publishing and access settings show that AI-generated apps need the same security reviews as conventionally developed software.
13-year-old asked an AI about suicide on her school laptop for weeks without triggering an alert
11.11.2026
The parents of a 13-year-old girl accuse a school monitoring system of taking weeks to detect conversations with Gemini about suicidal intentions. The case highlights the limits of automated safety systems and the responsibility of schools and AI providers in dealing with minors.
US Congress plans to question Nvidia CEO Jensen Huang over alleged espionage case
11.11.2026
US lawmakers want to question Nvidia CEO Jensen Huang about his connections to an executive suspected of spying for China. The case shows that geopolitical risks and scrutiny of personal connections are increasingly becoming part of corporate governance in the AI sector.
🛠️ Tools & Products
Gemini Decisions: Google launches a specialized AI model for rapid decisions instead of text generation
11.11.2026
Google DeepMind is introducing an API for clearly defined decisions such as yes-or-no questions, multiple-choice questions, and ratings. According to the company, it can also process images, audio, video, and PDFs. For developers, the interface could provide large numbers of structured decisions more quickly than traditional text generation, along with confidence scores.
Databricks’ new AI agent tells CEO Ali Ghodsi the unvarnished truth
11.11.2026
Databricks has updated Genie so that the agent can handle multistep tasks over longer periods using a company-specific ontology. The move aims to equip AI agents with the business context they need to carry out internal analyses and follow-up tasks more independently.
Google’s Tabletop experiment lets users build their own collaboration apps
11.11.2026
Google’s experimental Tabletop creates collaborative apps from descriptions and templates, such as trip planners, polls, or expense-sharing tools. The approach lowers the barrier to creating custom tools, but it remains unclear how well such AI-generated workspaces can be controlled and integrated.
DeepL Voice now transfers tone and speaking pace to other languages with a new voice model
11.11.2026
With its new Aure model, DeepL Voice aims to carry tone, emotion, emphasis, and speaking pace into real-time translations. This could make translated conversations sound more natural and better preserve nuances such as changes in mood or sarcasm.
Adobe’s AI experiments let users walk through photos and interview characters
11.11.2026
Adobe is showcasing experiments that turn photos into explorable 3D scenes, enable conversations with AI characters, and simplify image editing. The prototypes point to new creative workflows, but are not yet announced as standard features in Adobe products.
Gilbert + Tobin is rethinking the law firm business with AI
12.11.2026
OpenAI has published a conversation with Sam Nickless, CEO of law firm Gilbert + Tobin, about using AI in legal work. The account offers insight into how a law firm is rethinking its ways of working and its service model as it adopts AI.
ChatGPT Plus for veterans and members of the armed forces
11.11.2026
OpenAI reports that more than 20,000 veterans and people transitioning out of military service have claimed a free year of ChatGPT Plus. The program shows how AI providers use targeted offers to encourage practical use cases such as career planning.
🤖 Robotics
Serve Robotics opens its robot delivery network to external developers
11.11.2026
Serve Robotics is opening its delivery network to external developers, who will be able to build on it. A more accessible platform could enable more applications and partnerships around autonomous delivery, rather than limiting use to a single provider.
Tronix3D invests $4 million to expand its Pittsburgh site and create 47 jobs
11.11.2026
Tronix3D is investing $4 million to expand its operations in Pittsburgh and plans to create 47 new jobs. The expansion is a concrete sign of growing demand for 3D printing capacity and the expansion of regional manufacturing.
“Made in India” robots for the global market: Kody Robots to showcase at NRF 2027
11.11.2026
Kody Robots plans to showcase its AI-powered advertising robot Odigo at NRF 2027 in New York. The appearance brings an Indian robotics product to an international retail trade show and highlights its use in brick-and-mortar retail.
Study of an improved RRT-APF algorithm for path planning in tomato-harvesting robots
11.11.2026
A study combines RRT path planning with artificial potential fields and curve optimization to plan the movement of tomato-harvesting robots more efficiently. In simple obstacle scenarios, average planning time fell considerably compared with several benchmark methods, which is relevant to faster motion planning in agricultural robotics.
Simulation-based evaluation of a modified RRT algorithm for ground robot navigation in 3D environments
11.11.2026
Researchers are evaluating a modified RRT algorithm for ground robot navigation in 3D simulations using ROS 2 and Gazebo. Tests involving dynamic obstacles, changing goals, and difficult terrain are intended to show how robust the planning is beyond simplified 2D maps.
Clinically prioritized task management and verified safety oversight for a physician-supervised ward robot
11.11.2026
A simulation study describes a task-management system for ward robots that takes urgency, waiting times, and safe interruption points into account. The model addresses a central requirement for healthcare deployment: coordinating multiple clinical tasks without bypassing human approvals or non-interruptible processes.
Quantitative fall threshold for graded unilateral attenuation of motor commands in simulated humanoid walking control
11.11.2026
The study examines in a physics simulation how unilateral attenuation of motor commands affects the stability of eight humanoid walking controllers. Such stress tests provide indications of how robust learned motion controllers are to partial actuator or control failures without retraining the model.