Sony AI’s autonomous robot ‘Ace’ has reached expert-level performance in competitive table tennis, according to a study published in Nature on April 22, 2026. Built around an eight-jointed robotic arm and a high-speed sensing suite (multiple high-frame-rate cameras and event-based vision systems), Ace uses model-free deep reinforcement learning trained in simulation to perceive spin, predict trajectories and execute shots with millisecond latency. In April 2025 it beat three of five elite players under International Table Tennis Federation rules; after further development it recorded victories over professional opponents in December 2025 and March 2026, including a win against a top-25 player. Matches were officiated by licensed umpires on regulation courts. Researchers say Ace’s strengths include superior spin estimation, rapid reaction time and unpredictability; limitations include reliance on an external camera network, a non-humanoid mounting platform and limited ability to read human body language. Sony AI and outside experts highlight potential applications of the underlying perception and control technologies in manufacturing, service robotics, sports training and other fast, safety-critical environments, while also noting ethical and security considerations as capabilities mature.

Sony AI’s autonomous robot Ace has reached expert-level play in table tennis, competing under International Table Tennis Federation rules at an Olympic-sized court in Tokyo and sometimes defeating top human opponents, according to a study published in Nature on April 22, 2026. Built around an eight-jointed arm and a high-speed vision suite (nine cameras plus three gaze systems), Ace uses model-free reinforcement learning trained in simulation and proprietary control software to track spin, predict trajectories and react in about 20 milliseconds. In April 2025 it beat three of five elite players (who train ~20 hours weekly) and initially lost to two professionals; Sony AI says the robot later beat professional players in December 2025 and again in March 2026, including matches against top-ranked competitors such as Miyuu Kihara. Licensed umpires officiated matches and researchers emphasise parity constraints to avoid unfair hardware advantages. The team says the methods could transfer to manufacturing, service robotics, entertainment and other safety-critical, high-speed human-robot interaction tasks.

Meta has begun installing a tool called the Model Capability Initiative (MCI) on U.S.-based employees’ work computers to capture mouse movements, clicks, keystrokes and occasional screenshots for use as training data for its artificial intelligence models, Reuters and other outlets reported on April 21-22, 2026. The software runs on a curated list of work-related apps and websites and is presented as part of an internal push — rebranded as the Agent Transformation Accelerator — to build AI agents that can navigate interfaces and automate routine tasks. Meta says the data will be used only for model training, will not be used for performance reviews and that safeguards are in place to protect sensitive content, though it has not detailed exclusions. The move comes as Meta ramps AI spending and reorganises engineering teams, while also planning roughly 10% global workforce reductions beginning in May, prompting employee concern and wider criticism from privacy advocates and labour experts. Legal experts note the approach faces fewer hurdles in the U.S. than under EU data-protection rules.

A UK Competition Appeal Tribunal has certified a collective opt-out lawsuit against Microsoft alleging it overcharged nearly 60,000 British businesses for Windows Server licences when run on rival cloud platforms. The claim, brought by competition lawyer Dr Maria Luisa Stasi, seeks about £2.1 billion (roughly $2.8 billion) and contends Microsoft set higher wholesale prices for Windows Server on Amazon Web Services, Google Cloud and Alibaba Cloud than for equivalent usage on its own Azure service, making competitors comparatively more expensive. The tribunal dismissed Microsoft’s challenge that the claim lacked a workable method for calculating losses and concluded the action “comfortably crosses” the threshold for a real prospect of success. A Collective Proceedings Order will be published with opt-out details and timelines; Microsoft said it would appeal and denied any wrongdoing. The ruling comes amid growing regulatory scrutiny of Microsoft’s cloud and licensing practices by the UK’s Competition and Markets Authority and probes in Europe, the US and elsewhere. The decision keeps a major test of cloud licensing and competitive practices on track for trial and could lead to damages, settlements or changes to licensing terms if claimants prevail.

Apple on April 22 released iOS 26.4.2 and iPadOS 26.4.2, delivering targeted bug fixes and security updates for iPhone and iPad users. Apple’s release notes cite fixes to Notification Services — including a logging issue with improved data redaction and a remedy for notifications marked for deletion that could be retained unexpectedly. The update is available via Settings → General → Software Update. The release follows iOS 26.4 and last month’s 26.4.1, which enabled Stolen Device Protection and addressed iCloud sync problems; Apple also published a similar 18.7.8 update for older iOS 18 devices with the same security correction. No consumer-facing features are included in 26.4.2; feature additions are expected in the upcoming iOS 26.5 beta and the next major iOS 27 announcement at WWDC, where Apple may introduce items such as end-to-end encrypted RCS messaging and Maps changes. Apple classifies the patch as routine security maintenance rather than an emergency background security push.

Meta Platforms said on April 21-22, 2026 it has begun construction of a data centre in Tulsa, Oklahoma, a project valued at more than $1 billion located at Fair Oaks Innovation Park. The campus will span more than 2 million square feet, be built by Fortis Construction, and will be Meta’s 28th U.S. facility and 32nd worldwide. The company expects peak construction to create over 1,000 jobs and about 100 permanent operational roles, and has pledged to pay site water and wastewater costs and more than $25 million for local infrastructure upgrades. Separately, Meta disclosed a ‘Model Capability Initiative’ that deploys tracking software on U.S. employees’ computers to record mouse movements, clicks, keystrokes and intermittent screenshots to train AI models; Meta says the data won’t be used for performance reviews. The announcements come as Meta accelerates AI investment amid reports of planned layoffs and rising environmental and privacy scrutiny.