IBM and the United States Tennis Association have added AI serve analysis and match explanations to the 2026 US Open. Serve Quality will cover all 254 singles matches, tracking 21 points on a player’s body and racquet 50 times per second. IBM expects the tournament to generate about 1.2 billion movement points.
The feature turns those points into ratings for elements such as ball speed, placement, motion, and racquet movement. Fans can also use Key Moments, personalized updates, and Match Chat in the US Open app and website. The measurement scale is documented; the independent accuracy of the resulting scores and explanations is not. That distinction matters when a precise-looking number becomes part of how people understand athletic performance.
A hearing aid divides listening among four neural networks
Signia launched MaX, a hearing-aid platform that assigns speech, background noise, the surrounding soundscape, and the wearer’s own voice to four specialized neural networks. The first Pure C&G MaX device is available in the United States, Germany, and Nordic markets.
Signia says 97% of wearers in its testing performed better in noisy group conversation, and reports substantial gains in processing and energy efficiency over Signia IX. The announcement does not give participant counts or full protocols. These are manufacturer-authored results, not an independent clinical-outcome study.
Shoppers say they are more ready for AI than retailers
Narvar surveyed 1,348 U.S. consumers and 100 retail decision-makers for its 2026 Holiday Shopping Report. Sixty-five percent of consumers said they plan to use AI for holiday shopping: 43% for gift discovery, 33% for comparisons and review summaries, and 29% for budgeting. Only 8% of retailers described themselves as very confident in their readiness.
The gap is easy to picture, but it measures intention, not observed shopping behavior. The release also omits recruitment details and questionnaire wording. Narvar sells retail technology, so the sponsor’s commercial interest belongs beside the numbers.
AWS proposes an address book for agents and tools
AWS introduced Agent Registry and described Agentic Resource Discovery, an Apache-2.0 specification for finding agents, MCP servers, skills, APIs, and tools across separate catalogs. Records can carry approval, deprecation, and authorization information instead of relying on a single public list.
The specification and reference implementation are concrete. Broad interoperability is not: the announcement does not yet show independent vendors adopting the protocol in production. For now, ARD is a proposal for the permission-controlled address book that increasingly capable assistants will need.
Thomson Reuters builds a model around its professional archive
Thomson Reuters launched Thomson, its first internally controlled large language model for legal and professional work. The company says it invested $40 million in training and gave legal and AI academics evaluation access.
The release does not publish model weights, a full benchmark suite, evaluator findings, or a cost comparison. “Frontier” is the company’s description. The durable change is strategic: a professional publisher with a large trusted archive is choosing to own a specialist model rather than rent every answer from a general-purpose chatbot.
Live translation moves into a wearer’s field of view
Timekettle’s Babel OS 3.0 now powers streaming translation displayed inside RayNeo iO smart glasses. The companies say the system supports 55 languages and 109 accents, allowing a wearer to read translated speech while looking at the person speaking.
No public benchmark in the announcement covers latency, word error, accent-by-accent performance, privacy, or battery life. The integration verifies a new interface, not dependable translation in every conversation.
The UK and Ukraine sign a non-binding defense AI declaration
The UK and Ukrainian governments signed a declaration covering co-developed models, secure data and computing, joint assurance, industry deployment, cyber security, synthetic data, and research on autonomy.
The document is political intent, not an operational result. It is explicitly non-binding and gives no program budget, named deployed system, or enforceable safeguard. Even so, it shows AI policy moving beyond commercial chatbots into military autonomy, data sovereignty, and allied assurance.
Clinic conversation becomes a prescription draft
First Databank and Tebra announced the first commercial deployment of Script Agent. It turns ambient clinic conversation into a structured prescription draft grounded in medication terminology, then holds the order for clinician review.
The product does not authorize medication on its own. The announcement provides no patient sample, error rate, privacy assessment, or clinical-validation result. Moving from visit notes to an action draft makes those missing safety measures more consequential.
South Australia reports A$6 million in scam losses
South Australia’s government reported more than A$6 million in scam losses during the first half of 2026, including over A$3.5 million in investment scams and A$730,000 in account and identity-takeover scams. Officials warned that criminals are combining AI with stolen personal information.
The source does not say what share of the A$6 million involved AI. The loss figures make the risk concrete, but they do not support the stronger claim that AI caused all of it.
Groq 3 LPX enters production for Vera Rubin
NVIDIA says Groq 3 LPX is in full production as part of its Vera Rubin platform, with Nebius as the first announced adopter. The company reports 3,400 output tokens per second on Gemma 4 31B with a 100,000-token context, four times its named nearest alternative.
That speed result uses a vendor-selected model, context, and comparison. Independent cost, power, and end-user latency results are not included. Production status and the named adopter are verifiable; the broader performance advantage remains a company claim.
This article was written with assistance from Wonder Bricks AI Agent and edited by SunnyLabs.