NerdNews

Microsoft and OpenAI have revised their partnership agreement, introducing an independent expert panel to verify the arrival of artificial general intelligence (AGI). The deal extends Microsoft's IP rights until 2032 or until AGI is achieved, and allows OpenAI to develop products with third parties. The revenue-sharing arrangement will continue until AGI is verified, with payments extending over a longer period.

GitHub is launching a hub called Agent HQ, allowing developers to access and manage multiple AI coding agents, including OpenAI's Codex and Anthropic's Claude, alongside GitHub Copilot. This new feature enables developers to control and track AI agents, run them in parallel, and choose the best output. GitHub is also introducing a Plan Mode in VS Code and a code review step to Copilot.

OpenAI has completed its for-profit restructuring and struck a new deal with Microsoft, which includes a clearer plan for achieving artificial general intelligence (AGI). The deal also extends Microsoft's IP rights to OpenAI's technology through 2032 and allows OpenAI to collaborate with third parties on certain products.

Adobe launches Firefly Image 5 with support for layers, allowing creators to make custom models. It can work at native resolutions of up to 4 megapixels and enables layered and prompt-based editing. The update also includes new features such as generating speech and soundtracks, and a redesigned video generation and editing tool.

Elon Musk's Grokipedia has launched with around 885,279 articles, many of which are directly derived from Wikipedia pages. The site relies heavily on AI-generated content, which has raised concerns about bias and accuracy. Even Musk's own AI chatbot, Grok, has questioned the trustworthiness of Grokipedia.

Nephrogen, a biotech startup, combines AI and gene therapy to reverse kidney disease. The company's founder, Demetri Maxim, was motivated by his own experience with Polycystic Kidney Disease (PKD). Nephrogen has developed a delivery mechanism that is 100 times more efficient than current FDA-approved methods. The company plans to advance its novel delivery mechanism into clinical studies in 2027 and is raising a $4 million seed round.

The US Department of Energy is working with AMD to build sovereign AI supercomputers, including Lux and Discovery, which will be powered by AMD Instinct MI355X GPUs, AMD EPYC CPUs, and other technologies. The supercomputers will help researchers tackle challenges in energy, medicine, and national security, with Lux scheduled to deploy in 2026 and Discovery in 2029.

Nvidia is betting on an AI-driven robot workforce, partnering with companies to build robotic factories, autonomous collaborative robots, and humanoid robots for household chores. The company aims to address labor shortages and improve productivity, with its vice president stating that robots could account for over half a million open manufacturing jobs.

OpenAI released estimates that around 0.07% of active ChatGPT users show signs of mental health emergencies related to psychosis or mania, and 0.15% have conversations that include explicit indicators of potential suicidal planning or intent. The company has updated GPT-5 to respond more effectively to these situations, with the goal of directing users to professional help.

Waabi, a self-driving truck startup, has unveiled a new autonomous truck made in partnership with Volvo. The truck, called the Volvo VNL Autonomous, features Waabi's tech, including its sensor suite, compute, and the Waabi Driver software. Waabi's CEO, Raquel Urtasun, stated that the company has the potential to be the first to commercialize self-driving trucks without a human safety driver or observer.

Stellantis will build the vehicles, Nvidia and Foxconn will provide the self-driving tech, and Uber will deploy them on its platform, with production starting in 2028 and 5,000 vehicles initially deployed in the US

Nvidia and Uber are partnering to build a fleet of 100,000 fully self-driving robotaxis, with the vehicles set to start rolling out in 2027. The fleet will utilize Nvidia's Drive AGX Hyperion 10 in-vehicle computer, enabling level-4 automation and allowing the vehicles to drive themselves without human intervention.

Researchers have found that several AI browsers, including OpenAI's Atlas, are vulnerable to prompt injection attacks, which can lead to data exfiltration and other malicious activities. The attacks occur when unwanted text is entered into the browser's prompt input, allowing hackers to manipulate the AI into performing unintended actions.

Qualcomm has developed two new chips, the A1200 and A1250, based on its existing neural processing unit (NPU) technology, to power AI inference workloads. The A1250 features a novel memory architecture for near-memory computing, providing a generational leap in efficiency and performance. The chips are designed to deliver low total cost of ownership and are optimized for large language model and multimodal model inference.

The US military is testing AI-powered drones controlled by large language models, with the goal of streamlining kill chains and making them more efficient. Companies like Anduril and Anthropic are working on AI-related military contracts, with the Pentagon planning to increase spending on AI in the coming years.

Researchers developed a side-channel attack called TEE.Fail, which extracts secrets from the trusted execution environment in Intel, AMD, and NVIDIA CPUs. The attack requires physical access and root-level privileges, but can be done with less than $1,000 in hardware. It exploits weaknesses in DDR5 memory encryption, allowing attackers to recover private signing keys and forge valid TEE quotes.

Amazon has announced a reduction of approximately 14,000 positions in its corporate workforce, citing the need to adapt to the rapidly changing world of AI. The impacted departments include video games, logistics, payments, and cloud-computing. This move is part of Amazon's effort to become more lean and agile in order to innovate faster and better serve its customers.

Google's smart home ecosystem is getting a major upgrade with Gemini for Home, a next-gen smart home assistant that uses Google's large language model. Gemini promises to be more advanced and less frustrating than the current Google Assistant, with features like automation assistance and advanced camera notifications. However, Google is asking for money for premium plans, and it remains to be seen if the company can deliver on its promises.

A critical zero-day flaw in Google Chrome, tracked as CVE-2025-2783, has been exploited in the wild as part of a targeted espionage campaign. The attacks, linked to the group known as Mem3nt0 mori, involved tools developed by the Italian spyware vendor Memento Labs. The flaw stemmed from a logical oversight in Windows' handling of pseudo handles, allowing attackers to execute code in Chrome's browser process.

Elloe AI is a platform that adds a layer to companies' LLMs to check for bias, hallucinations, errors, compliance issues, misinformation, and unsafe outputs. It uses an API or SDK that sits on top of an AI model's output layer and fact-checks every single response against verifiable sources, regulations, and audit trails.