ZKBoost Protocol Unveiled for Private, Verifiable XGBoost AI Model Training

Researchers have introduced ZKBoost, a groundbreaking cryptographic protocol designed to verify XGBoost machine learning model training on sensitive datasets without revealing the underlying data or model parameters. ZKBoost is the first zero-knowledge proof of training (zkPoT) protocol specifically developed for XGBoost, a widely-used algorithm for tabular data. This innovation addresses the critical need for cryptographic guarantees of model integrity in high-stakes applications. The system employs a fixed-point implementation of XGBoost compatible with arithmetic circuits, enabling efficient zero-knowledge proofs. Research indicates this implementation maintains accuracy within 1% of the standard XGBoost, making practical verification on real-world data a reality.

Sources:

Major Data Breach: ‘Chat & Ask AI’ App Leaks 300M Messages via Firebase Flaw

A significant data breach has exposed 300 million private messages belonging to 25 million users of the ‘Chat & Ask AI’ application. The leak stemmed from a known Firebase misconfiguration, which left the application’s database publicly accessible without requiring authentication. The exposed data included users’ complete chat histories, their preferred AI models, and other personal settings. The ‘Chat & Ask AI’ app serves as a wrapper, offering access to various large language models such as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini. The developer, Codeway, reportedly secured the database within hours of being notified by the security researcher who discovered the vulnerability.

Sources:

Trener Robotics Raises $32M Series A to Advance AI Skills Platform for Manufacturing Robots

Trener Robotics, formerly T-ROBOTICS, has successfully closed a $32 million Series A funding round to enhance its AI-powered skills platform for manufacturing robots. The company’s Acteris platform enables operators to automate complex tasks by describing them in natural language, which the system translates into executable robot actions. This innovative approach uses physical AI to master vision, language, and movement, allowing robots to adapt to dynamic production environments in real-time. The new capital will accelerate R&D, train new robotic skills, expand the team, and grow its market and partnerships, bringing Trener Robotics’ total funding to over $38 million.

Waymo Launches Fully Autonomous, Driverless Ride-Hailing Service in Nashville

Alphabet subsidiary Waymo has officially launched its fully autonomous ride-hailing service in Nashville, Tennessee, completely removing human safety drivers from its vehicles. After months of testing in the city, the company is now proceeding with a commercial launch, which includes a partnership with Lyft. Initially, customers can book rides through the Waymo app, with availability on the Lyft app to follow. This expansion signifies another major step in the escalating competition within the U.S. robotaxi market. Waymo currently operates in several other major cities, including Atlanta, Austin, Los Angeles, Miami, the San Francisco Bay Area, and Phoenix.

Sources:

US Military Warns Bomb Techs Against Using Generative AI for Sensitive EOD Data

The U.S. military’s primary authority on Explosive Ordnance Disposal (EOD) technology has issued a formal warning to bomb technicians regarding the risks of uploading sensitive technical material into generative AI systems. The warning specifically highlighted the Automated Explosive Ordnance Disposal Publication System (AEODPS), a critical database with classified information on bomb disposal. While no known incidents have occurred, the advisory was posted to the Joint Explosive Ordnance Disposal (JEOD) portal as a proactive security measure. The core concern is that uploading such data, even to Pentagon-approved platforms, could create significant national security risks if compromised.

Kubernetes Cluster API v1.12.0 Released to Simplify Multi-Cluster Management

The Kubernetes community has released Cluster API v1.12.0, a major update designed to streamline the provisioning, upgrading, and operation of multiple Kubernetes clusters. This version introduces key features like ‘In-place Updates,’ which allows platform engineers to modify machine details without disrupting active workloads. Another significant enhancement is ‘Chained Upgrades,’ enabling administrators to upgrade across multiple versions in a single operation, drastically reducing manual effort. These improvements aim to lower the barrier to entry for maintaining secure and up-to-date Kubernetes environments by automating complex lifecycle management tasks.

Sources:

Nu Quantum Opens Cambridge Lab for Trapped-Ion Quantum Networking

Nu Quantum has inaugurated a new laboratory in Cambridge, UK, dedicated to trapped-ion networking. This facility is the first industrial R&D center in Europe focused on distributed trapped-ion quantum computing. The lab will be the primary testbed for the company’s Entanglement Fabric roadmap, which aims to connect clusters of commercial quantum processors. This expansion, doubling their research infrastructure, follows a recent $60 million Series A funding round. The new facility integrates advanced optical microcavity technology to improve qubit-photon connections and features a specialized laser suite developed with the National Quantum Computing Centre (NQCC).

Google Quantum AI Achieves Critical Breakthrough in Error Correction

The Google Quantum AI team has reported a significant advancement in quantum error correction, a crucial step toward building large-scale, fault-tolerant quantum computers. In new experiments, researchers demonstrated that by bundling more physical qubits together to form a single logical qubit, they could reduce the error rate below a critical threshold. This breakthrough is fundamental to creating quantum machines capable of solving practical, real-world problems. Analysts believe fault-tolerant quantum hardware could accelerate technological development, including solutions for reducing carbon emissions, though the energy consumption of cooling systems remains a challenge.

Sources:

Infleqtion Awarded $6.2M ARPA-E Contract for Quantum-Powered Grid Optimization

Quantum computing company Infleqtion has executed a $6.2 million contract with the U.S. Department of Energy’s ARPA-E to develop quantum solutions for energy grid optimization. The project, named ENCODE (Enhancing Neutral-atom Computers for Optimizing Delivery of Energy), will apply quantum-enhanced computational methods to solve complex optimization challenges facing grid operators. These challenges are driven by rising electricity demand from electrification and AI computing. Infleqtion will utilize its neutral-atom processors and Superstaq optimization layer in collaboration with Argonne National Laboratory, NRL, EPRI, and ComEd to improve energy affordability and grid stability.

Sources:

Salesforce Moves Heroku to Maintenance Mode, Halts New Enterprise Contracts

Salesforce is shifting its strategy for the Heroku platform-as-a-service, moving it into a “sustaining engineering” phase. This means prioritizing stability, security, and reliability over the development of new features. As part of this change, the company will no longer offer new Heroku Enterprise contracts. Existing customers, both on enterprise agreements and credit-card plans, can continue to use and renew their services with no immediate changes to pricing or support. Industry analysts view this as a strategic deprioritization of Heroku, suggesting a long-term managed decline for the influential PaaS as Salesforce deepens its focus on AI-driven platforms.

Sources:

JDK 26 Finalizes Feature Set with Focus on Pattern Matching and Performance

The feature set for JDK 26 has been finalized, targeting ten new features for the upcoming release. A key enhancement is the fourth preview of primitive types in patterns, instanceof, and switch, which aims to allow primitive types in all pattern contexts. Other notable additions include ahead-of-time object caching to improve startup performance, a second preview of lazy constants, and HTTP/3 support in the Client API. The release also emphasizes performance gains, such as reducing synchronization in the G1 garbage collector to boost throughput. Early-access builds are now available for developers.

Sources: