NerdNews

Google has rebranded and expanded its Vertex AI developer platform into the Gemini Enterprise Agent Platform, aiming to provide a single system for developing, deploying, governing, and monitoring AI agents across organizations. The platform includes features such as Agent Studio, Agent Runtime, and governance capabilities like Agent Identity and Agent Gateway. Google also announced infrastructure updates, including its eighth generation of TPU chips and security updates through its purchase of Wiz.

Google has unveiled its 8th-gen tensor processing units (TPUs), with two new AI accelerators: TPU 8t for training and TPU 8i for inference. The TPU 8t is designed for large-scale training, with 216 GB of high-bandwidth memory and up to 12.6 petaFLOPS of 4-bit floating point compute. The TPU 8i is optimized for inference, with a larger SRAM cache and faster memory pool. Both chips will be available in Google Cloud Platform later this year.

X has introduced Grok-powered Custom Timelines, allowing users to create personalized feeds on over 75 topics. The feature uses AI to curate and personalize content for individual users, and is currently available to Premium subscribers on iOS.

Google is introducing 'auto browse' agentic capabilities to Chrome, allowing users to automate tasks like booking travel and inputting data. The feature uses Gemini to understand live context in open browser tabs and can be enabled via policy. Google is also expanding its security measures to detect unsanctioned AI tools in the workplace.

Thinking Machines Lab, founded by former OpenAI executive Mira Murati, has signed a new multibillion-dollar agreement with Google Cloud to expand its use of AI infrastructure, including systems powered by Nvidia's latest GPUs. The deal is valued in the single-digit billions and includes access to Google's latest AI systems and infrastructure services.

Cloudflare announces general availability of Sandboxes, providing persistent, isolated Linux environments for AI agent workloads with secure credential injection, PTY terminal support, and active CPU pricing.

Cloudflare has outlined a reference architecture for scaling Model Context Protocol (MCP) deployments, focusing on centralized governance, remote server infrastructure, and cost controls to address security and governance risks in enterprise MCP deployments.

Microsoft released an emergency patch for ASP.NET Core to fix a high-severity vulnerability that allows unauthenticated attackers to gain SYSTEM privileges on devices that use the Web development framework to run Linux or macOS apps. The vulnerability, tracked as CVE-2026-40372, affects versions 10.0.0 through 10.0.6 of the Microsoft.AspNetCore.DataProtection NuGet package.

Mozilla's CTO, Bobby Holley, believes that Anthropic's Mythos AI model can help defenders gain the upper hand in security. The model found 271 vulnerabilities in Firefox 150, but none that a human couldn't spot. Holley thinks this technology can help close the gap between machine-discoverable and human-discoverable bugs, making it cheaper for defenders to find vulnerabilities.

A new supply chain attack targeting the Node Package Manager (npm) ecosystem is stealing developer credentials and attempting to spread through packages published from compromised accounts. The attack was spotted in multiple packages from Namastex Labs and uses techniques similar to TeamPCP's CanisterWorm attacks. Developers should remove affected packages and rotate potentially exposed secrets.

ChatGPT's new Images 2.0 model can generate complex images, including text, with high fidelity and specificity. It has 'thinking capabilities' that allow it to search the web, create multiple images from one prompt, and double-check its creations. The model can generate images in various sizes and resolutions, including multi-paneled comic strips and marketing assets.

Google has announced the release of new AI security agents to fight threats, along with tools to control and secure these agents. The company's strategy is to use AI to fight AI, with a focus on machine speed and autonomy. The new agents include a Threat Hunting agent, a Detection Engineering agent, and a Third-Party Context agent, all designed to work together to identify and mitigate security threats.

Google has unveiled two new Tensor Processing Units (TPUs), the TPU 8t for training and the TPU 8i for inference, designed to power its Gemini-based agents and support third-party developers. The new TPUs offer improved performance, efficiency, and scalability, with the TPU 8t capable of reducing training time for frontier AI models from months to weeks.

Google Cloud announced its eighth generation of custom-built AI chips, with the TPU 8t for model training and TPU 8i for inference, offering up to 3x faster AI model training and 80% better performance per dollar. The company will continue to use Nvidia-based systems in its infrastructure, but may eventually reduce its reliance on Nvidia as it grows its AI cloud business.

OpenAI has partnered with Infosys to integrate its AI tools into the Indian IT giant's Topaz AI platform, aiming to help clients modernize software development and deploy AI systems at scale.

A new report reveals that 11 data centers, powered by gas power plants and generators, could emit over 129 million tons of greenhouse gases per year, exceeding the emissions of entire nations like Morocco. This contradicts the net-zero emissions pledges made by these companies, highlighting the need for sustainable practices in the tech industry.

The article discusses the concept of sovereign fault domains, which are failure boundaries defined by legal, political, or physical jurisdiction. It highlights the need for architects to consider these domains when designing cloud systems, as traditional high-availability models may not be sufficient. The article provides examples of geopolitical events that can compromise cloud regions and discusses design patterns for sovereign resilience, including multi-region architecture, jurisdiction-aware data abstraction, and replication within sovereignty.

Database vendors are using Large Language Models (LLMs) to build natural language query systems, aiming to make data querying more accessible. However, experts warn that these systems can produce syntactically correct but semantically wrong queries, and human oversight is necessary to ensure accuracy.

Lloyds Register has evaluated an AI-based navigation system for shipping, using a computer vision system to identify and categorize complex navigation scenarios. The system, called Orca AI SeaPod, was tested on a container ship and demonstrated its ability to detect close-range and low-signature targets. The trial aimed to assess the performance of the system and its potential to enhance situational awareness and support human decision-making at sea.

Elon Musk announced that millions of Tesla owners will need hardware upgrades to run a future version of its Full Self-Driving software. The upgrades will include a new computer and cameras for cars with Hardware 3, which was used in Teslas sold between 2019 and 2023. Musk stated that Hardware 3 is insufficient to run more advanced versions of Full Self-Driving, and the company is planning to build 'micro-factories' to handle the upgrades.