Nvidia’s Leadership Shake-Up Makes Governance an Industry Risk
In Brief
- Nvidia’s leadership shake-up makes governance an industry risk: CFO Colette Kress is temporarily leading the key AI chip supplier on an interim basis. Questions about succession, export controls, and open source remain unresolved.
- AI agents are moving closer to real tasks—and real risks: New features take on calls, meetings, and marketing work. Security concerns and data leaks underscore the importance of clear access boundaries.
- Google opts for the right model rather than always choosing the biggest: Gemini 4 Hydrogen is designed to handle complex requests quickly and at lower cost, and is being rolled out across Search, apps, and developer offerings.
- Regulation and robotics are becoming more concrete: Australia plans mandatory evaluations for frontier AI, while new gripping and tactile sensors are intended to make robots more practical to deploy.
🔬 Models & Research
Google’s AlphaGenome 2 aims to show how every body cell reads its DNA
12.11.2026
Google DeepMind has unveiled AlphaGenome 2. The model is designed to predict genetic changes and DNA regulation in specific human cell types. Cell-type-specific analysis could help pinpoint relevant genetic variants. The model is initially available for non-commercial research.
Why AI models can hide their thoughts—and why countermeasures barely work
12.11.2026
An OpenAI study describes training steps that lead powerful models to reveal less of their chain-of-thought traces. For safety oversight of AI agents, this means that visible reasoning is not a reliable verification mechanism. Transparency also depends on trade-offs involving performance and cost.
Study: AI chatbots recommend fake online stores
12.11.2026
A test involving 151 brands found that ChatGPT, Gemini, and Perplexity sometimes direct users to fake or risky websites when they search for products. Companies should therefore not equate AI recommendations with reliable brand discoverability, and should continue directing customers to safe, official sources.
No, AI agents did not discover a previously unknown Maya site
12.11.2026
A possible Maya discovery reported by AI agents had already been documented in an academic journal the previous year. The case shows that although agents can find patterns and clues, their findings can easily be misinterpreted without subject-matter expertise and a review of the academic literature.
Optimizing drone route planning with multiple weighted cost factors and improved Harris Hawks optimization
12.11.2026
The study combines flight distance, altitude, position, obstacle response, and distance from threats into a weighted objective function for drone route planning. This multi-objective approach could improve the safety of autonomous systems in hazardous environments, but it still needs to be validated under real flight conditions.
💼 Business & Markets
Jensen Huang has stepped down at Nvidia—who is interim successor Colette Kress?
12.11.2026
Nvidia co-founder Jensen Huang has stepped down as CEO, president, and board member, effective immediately. CFO Colette Kress is taking over on an interim basis. Since Huang shaped the company over several decades and had not prepared a successor, the transition is especially important for customers, partners, and the entire AI chip market.
Nebius tests the neocloud model with $40 billion in planned spending
12.11.2026
Neocloud provider Nebius is planning $40 billion in investments for 2027. The expansion is to be financed in part through customer prepayments and loans backed by contracts. Delays or project cancellations could put the model under pressure; the execution of its data center plans is the main risk.
Norwest deepens its push into AI with a new $4 billion fund
12.11.2026
Norwest has launched a new $4 billion fund and plans to increase investment in AI companies in areas such as healthcare, voice AI, agentic cybersecurity, and physical AI. The capital shows that investors are increasingly funding applications and infrastructure in established industries alongside major model providers.
SoftBank’s billion-dollar loss could curb the AI industry’s ambitions for natural gas
12.11.2026
SoftBank’s energy subsidiary reportedly reserved far more gas turbines than it needed for its planned energy projects, resulting in a substantial financial loss. The case highlights that the expansion of AI-driven data centers depends not only on chips and real estate, but also on risky energy and procurement planning.
More and more game developers are using generative AI
12.11.2026
According to the annual industry survey, 40 percent of game developers polled now use generative AI at work, up from 36 percent the previous year. Its broader adoption shows that AI tools are entering production workflows. However, their acceptance and their impact on creative work remain contentious.
Cyberattack on Fireworks AI gave hackers access to Notion customer chat transcripts
12.11.2026
An attack on Fireworks AI’s cloud environment enabled access to customer data. Notion says around one percent of its customers were affected. The incident shows that when using AI services, companies must include development environments and subcontractors in their security and privacy reviews.
Google DeepMind employees bring their union dispute to the UK government
12.11.2026
After talks with management failed to produce results, Google DeepMind employees are turning to the UK government to advance their organizing effort. The dispute shows that working conditions and employee representation are increasingly becoming part of corporate and talent strategies, even at leading AI labs.
⚖️ Policy & Regulation
Australia plans to require independent evaluations and safety checks for frontier AI companies
12.11.2026
Australia is planning mandatory safety rules for frontier AI providers. The proposal includes independent evaluations, ongoing monitoring, and a new oversight agency with the power to impose penalties. It could create additional evidentiary requirements for companies and make access to powerful models more dependent on verifiable safety processes.
Nvidia internally confirmed it can track chips to combat smuggling, while the industry publicly cast doubt
12.11.2026
According to a US lawmaker, Nvidia representatives confirmed that a chip’s location can be technically verified, even as an industry association publicly questioned whether this was feasible. The debate over possible geolocation requirements shows how export controls could directly shape hardware and infrastructure design in the future.
United Airlines blocks AI agents from booking flights, citing cybersecurity risks
12.11.2026
United Airlines plans to make it harder for AI agents to access bookings and reservations, citing cybersecurity risks and potential regulatory conflicts. Providers of autonomous assistants should expect companies to revisit identity verification, account access, and the limits of delegated actions.
Jensen Huang agrees to testify before the US Congress
12.11.2026
Former Nvidia chief Jensen Huang has agreed to testify before the US Congress on national security and chip-control issues. The hearing is likely to highlight political pressure over advanced AI chips and the possibility that they could be diverted to China.
🛠️ Tools & Products
Google introduces Gemini 4 Hydrogen, a fast and more affordable AI model for everyday use
12.11.2026
Google has released Gemini 4 Hydrogen for everyday programming tasks and agentic workflows. According to the report, the model offers a context window of one million tokens and several reasoning levels. As a faster, cheaper model for Search, apps, and the API, it is intended to help teams deploy capable AI more selectively according to task and cost.
Claude Meetings: Anthropic adds Claude as a participant in video conferences
12.11.2026
Claude Meetings can join video conferences, answer questions, and capture notes, tasks, and documents. The feature brings AI directly into team conversations and can reduce follow-up work. However, productive use requires clear rules on consent and data privacy.
Muse can now call businesses in the US—for a fee
12.11.2026
Meta’s AI assistant Muse can call businesses in the US for paying subscribers and handle tasks such as reservations or inquiries. This takes consumer AI beyond generating responses and into carrying out real-world actions. Transparent handoffs and reliable confirmation of user instructions are therefore especially important.
ElevenLabs Scribe v3 Live aims to stop interrupting callers mid-sentence
12.11.2026
ElevenLabs has introduced Scribe v3 for audio recognition and Scribe v3 Live for real-time conversations. The live version is designed to use conversational context so it does not interrupt speakers prematurely during pauses to think. For voice assistants and telephony applications, natural conversational flow now matters alongside recognition accuracy. According to tests, the higher costs and longer processing times must be weighed against the benefits.
Nectar launches an AI agent to automate creator discovery and outreach for brands
12.11.2026
Nectar has introduced a creator agent that finds suitable influencers, reviews content and brand partnerships, sends personalized messages, and tracks campaigns. This makes it possible to automate more of a marketing process that was previously manual. Brands still need to oversee quality, tone, and approvals.
Phota Labs uses AI to turn everyday photos into professional-looking shots
12.11.2026
The iOS app Phota Labs uses an AI photo assistant to automatically make everyday photos look more professional while preserving their character. The product focuses on automated editing rather than complex manual tools, targeting users who want better results without a time-consuming editing process.
🤖 Robotics
Eureka Robotics launches StereoPick for easy setup of 3D picking systems
12.11.2026
StereoPick is designed to set up 3D pick-and-place applications—from CAD model to operation—in a single tool, without programming, machine-vision expertise, or manual annotation. This lowers the barrier to flexible automation and could speed up the commissioning of robotic picking systems.
Melexis gives humanoid robots a sense of touch with its first tactile fingertip module
12.11.2026
Melexis has introduced its first tactile fingertip module for humanoid robots. Tactile sensing can add feedback to gripping tasks, making it an important component for more delicate manipulation that goes beyond vision-based perception alone.
From the assembly line to the test room: Align solves a complex transport challenge
12.11.2026
A manufacturer is using a low-profile, maneuverable Align transporter to move engine assemblies weighing up to 50,000 pounds and standing up to 14 feet tall through a low doorway. This eliminates the need to dismantle the engines beforehand. It can simplify internal transport routes and save time in the manufacturing process.