SSPAI Morning Brief: Microsoft Improves Windows 11 as Android Tightens App Sideloading Rules

Morning Brief

Microsoft promises multiple improvements to Windows 11
Android may require a “cooling-off” day before installing apps from outside official stores
Halide CEO accuses former co-founder poached by Apple
U.S. tech companies see rising competition in “token farming”
Meta’s metaverse plans move toward an end
Preprint platform arXiv to operate independently from Cornell University
News Worth a Quick Look

Microsoft promises multiple improvements to Windows 11

On March 20, Pavan Davuluri, Executive Vice President of Microsoft’s Windows and Devices division, announced that Windows 11 will receive major quality upgrades. The promised measures include—

The taskbar will once again support docking at the top, left, or right of the screen;
AI integration will be more restrained, with unnecessary Copilot shortcuts removed from core apps such as Screenshot, Photos, Widgets, and Notepad;
Users will be able to skip updates during initial setup on new devices, and in daily use can choose to shut down or restart without installing updates;
Core Windows experience components, including the Start menu, will be migrated to the WinUI3 framework to reduce interaction latency, with a focus on improving File Explorer launch and search performance;
The Widgets panel will adopt a quieter, less intrusive default design;
Stability improvements will be made to Windows Subsystem for Linux (WSL), Windows Hello biometric authentication, and peripheral connectivity.

As background, Windows 11 has recently pushed AI features heavily across its interface while also facing frequent stability issues, leading to user complaints. These optimizations will be rolled out for testing to Windows Insider Program members between March and April this year.

Android may require a “cooling-off” day before installing apps from outside official stores

On March 19, Google unveiled a new proposal targeting Android app “sideloading” (installing apps from non-official stores), which includes a so-called mandatory 24-hour cooling-off period.

According to the developer identity verification policy announced by Google last August, developers will be required to submit real names and identity verification documents to the platform. This mandatory verification will first be implemented in countries such as Brazil and Singapore starting this September, with a global rollout planned for 2027. The new sideloading process, including the cooling-off period, will be made available earlier in August as an optional setting for advanced users. The high threshold has drawn widespread criticism, and Google later promised in November to continue allowing sideloading for power users.

Under the newly outlined process, users who wish to enable sideloading must first activate Developer Mode, confirm they are not being coerced, reboot their device, and then wait 24 hours before verifying their identity via fingerprint or facial recognition. Once completed, users can choose to enable sideloading for either 7 days or indefinitely. Google states that the reboot step is intended to interrupt potential scammer activities such as ongoing calls or remote screen monitoring, while the cooling-off period is designed to break the sense of urgency often exploited by fraudsters, giving users time to think more carefully.

Halide CEO accuses former co-founder poached by Apple

According to The Information, after failing to acquire the developer behind the well-known photography app Halide, Apple instead recruited one of the startup’s co-founders. The move has since triggered a fierce legal dispute within the original team, with both sides accusing each other of misappropriating funds and stealing trade secrets.

The lawsuit was filed by Ben Sandofsky, CEO of Halide’s parent company Lux Optics, in the Superior Court of Santa Cruz County, California. He alleges that former co-founder and designer Sebastian de With misappropriated over $150,000 of company funds for personal expenses since 2022, and, after joining Apple, disclosed confidential materials and source code to the company. De With was dismissed in December last year following these allegations and officially joined Apple’s design team in January this year. In response, de With’s legal counsel strongly denied any intellectual property theft and counter-accused Sandofsky of financial misconduct, claiming the lawsuit is retaliation for de With’s request to review company financial records.

The dispute stems from acquisition talks between Apple and Lux Optics last summer. Apple had initially hoped to strengthen its native iOS camera app through the acquisition, explicitly stating that Lux’s intellectual property was a key consideration. However, the Lux team believed that future app updates could secure a better valuation, leading to the collapse of negotiations. Just two months later, Apple began recruiting de With independently. Apple is not named as a defendant in the case.

Sources indicate that upgrading the native camera app is currently a top priority within Apple. To enable the upcoming iPhone 18 Pro to rival professional-grade cameras in certain advanced features, Apple is actively seeking to absorb both the underlying technical expertise and design talent from specialized imaging apps.

U.S. tech companies see rising competition in “token farming”

According to The New York Times, major Silicon Valley companies such as Meta, Shopify, and OpenAI have recently begun incorporating AI usage into employee performance evaluations, sparking an internal competition among programmers dubbed “token farming” (or “Tokenmaxxing”). In an effort to demonstrate productivity to management, some employees are aggressively invoking AI tools, with individual usage reportedly exceeding $150,000 worth of compute in a single month. One OpenAI engineer processed as many as 210 billion tokens in just one week—equivalent to roughly 33 full Wikipedia texts.

Traditional prompt-and-response usage of AI is unlikely to reach such scale. Instead, employees are primarily relying on the rising class of agentic programming tools. These systems can operate autonomously around the clock without human supervision, spawning multiple sub-agents to handle tasks in parallel—continuing to consume vast amounts of compute even while the user sleeps.

According to reports, generous token budgets have effectively become a new form of employee perk, akin to free lunches. To encourage adoption of new technologies, companies have introduced internal leaderboards tracking AI consumption and even reward top users. This incentive structure has also given rise to opportunistic behavior.

Industry observers note that behind this frenzy of compute consumption lies a growing sense of “token anxiety” among white-collar workers. As AI becomes increasingly capable of independently writing code, programmers face the looming threat of replacement. To avoid falling behind in the wave of automation, employees are treating the operation of large-scale AI agent systems as a form of career self-defense—signaling to management that they are keeping pace with cutting-edge technology, even as the actual quality of code output takes a back seat.

Meta’s metaverse plans move toward an end

According to The New York Times, Meta has effectively brought an end to Mark Zuckerberg’s virtual reality (VR)-based metaverse vision. Last week, Meta announced that starting June 15, users will no longer be able to access its “metaverse social app” Horizon Worlds via VR headsets. Although Meta later clarified that it will continue to support some existing VR experiences within the app, it made clear that no new applications will be added.

Behind this retreat is a comprehensive shift of resources toward AI. The division responsible for metaverse projects has laid off around 10% of its staff in recent months. At the latest developer conference, Zuckerberg mentioned the “metaverse” only twice, compared to 23 mentions of “AI.” He is now betting on “superintelligence,” with Meta expected to invest at least $115 billion this year, primarily in AI development and large-scale data center infrastructure. Meanwhile, the company’s hardware focus has shifted to AR smart glasses equipped with built-in AI assistants.

Meta acquired VR headset company Oculus in 2014 for $2 billion, rebranded itself as Meta in 2021, and has accumulated approximately $80 billion in losses on its metaverse initiative. During the pandemic, the demand for remote work and contactless social interaction briefly propelled the metaverse concept into the spotlight. However, Meta’s virtual world repeatedly ran into technical limitations, with its crude floating avatars becoming a frequent subject of online ridicule. Additionally, the high cost of hardware has hindered broader adoption. Even Apple’s Vision Pro has yet to break into the mass consumer market, and VR has not replaced smartphones as the next computing platform.

Preprint platform arXiv to operate independently from Cornell University

According to Science, the globally renowned preprint platform arXiv has announced that it will officially separate from Cornell University on July 1 and become an independent nonprofit organization. The move aims to broaden funding channels in response to the surge in paper submissions in recent years and to invest resources in combating “AI slop.” arXiv allows scientists to share research findings quickly and freely before formal peer review, serving as a foundational platform for scholarly communication in fields such as physics, computer science, and mathematics.

Since its founding in 1991, the platform has grown rapidly, with the recent AI boom significantly accelerating submission volumes—expected to exceed 300,000 papers this year. To cope, the arXiv team has expanded to 27 members, driving its annual operating costs to $6.7 million and resulting in deficits over the past two years. Cornell officials stated that spinning off into an independent entity would alleviate concerns among potential donors about funds being controlled by the university, enabling arXiv to more directly attract financial support from libraries, research institutions, and foundations worldwide.

In response to the transition, some researchers have expressed concerns on social media, including criticism of the proposed CEO salary of up to $300,000 and worries about whether the platform might eventually be acquired by a commercial entity. Management responded that arXiv will maintain its successful nonprofit model, with no noticeable changes for users in the short term, and no immediate increase in institutional membership fees. Cornell University, the Simons Foundation, and others have already secured initial funding to ensure the new organization begins operations without a deficit.

Last year, bioRxiv and medRxiv—preprint platforms in biology and medicine inspired by arXiv—also separated from their academic hosts for similar reasons. As arXiv founder Paul Ginsparg has noted, while universities offer many advantages, historical experience suggests they are not well-suited to provide long-term, comprehensive support for global research infrastructure of this kind.

News Worth a Quick Look

According to PBX SCIENCE, Ubuntu 26.04 LTS, set to be released on April 23, will break a 46-year Linux tradition: when users enter their password with the sudo command, the terminal will display asterisks (*) to indicate input, instead of remaining blank. Since its introduction in 1980, sudo has provided no visual feedback during password entry, originally to prevent observers from inferring password length. In this update, Ubuntu adopts sudo-rs, a Rust-based reimplementation of sudo focused on security. The development team argues that modern graphical login interfaces already display password placeholders, making the terminal’s invisible input offer minimal security benefits, while often confusing newcomers into thinking the terminal has frozen or their keyboard is unresponsive.
According to The Wall Street Journal, OpenAI plans to integrate ChatGPT, Codex, and the Atlas browser into a single desktop “super app.” The move aims to streamline user experience, consolidate internal engineering resources, and refocus strategy on enterprise clients and developers. In an internal memo, OpenAI’s Head of Applications, Fidji Simo, noted that the team’s efforts are currently too fragmented, slowing development and making quality harder to maintain. The new app will emphasize AI agent capabilities. The existing ChatGPT mobile app will remain unchanged.
According to The Verge, Google is testing AI-generated rewrites of webpage titles in search results. The outlet found that titles of multiple articles were replaced with AI-generated versions without any indication of modification. A Google spokesperson confirmed this is a “limited” experiment using generative AI, intended to better match user intent and improve engagement. However, Google stated that if the feature is officially launched, it will not rely on generative models—without explaining how titles would then be rewritten. Google has previously adopted a similar approach in its Discovery feed.
According to Reuters, more than a decade after the failure of the Fire Phone, Amazon is planning a return to the smartphone market under the codename “Transformer.” The company aims to create a mobile device deeply integrated with AI and the Alexa voice assistant, focusing on AI-native features. Amazon is exploring multiple design directions, including minimalist “dumbphone”-style devices inspired by the Light Phone.
On March 21, the co-founder and COO of the AI coding app Cursor confirmed that its new model, Composer-2, is fine-tuned from Moonshot AI’s Kimi K2.5 model. Cursor released Composer-2 on March 19, claiming performance between OpenAI GPT-5.4 (high) and Anthropic Claude Opus 4.6 (high) at a lower cost. However, developers noticed the model ID in code as “Kimi-k2p5,” leading to speculation that Composer-2 was essentially a reinforced-learning version of K2.5. Cursor acknowledged that failing to mention Kimi as the base model in its launch blog was an oversight and said it would correct this in future releases. Kimi responded that Cursor accesses K2.5 via the FireworksAI inference and deployment platform under a licensed commercial partnership, and welcomed its integration into the broader open model ecosystem.
On March 21, China’s Cyberspace Administration stated that current standards for labeling short video content across platforms are inconsistent. Some videos involving fictional narratives, staged marketing, or AI-generated content lack proper labeling, misleading the public, disrupting social order, and polluting the online ecosystem. In response, regulators plan to standardize content labeling: first, by defining required label categories; second, by making labeling a mandatory step before publishing; and third, by retroactively reviewing and labeling existing content in phases. Authorities will soon issue detailed requirements and timelines, implement nationwide enforcement, and conduct inspections—penalizing accounts and platforms that fail to comply, with public disclosure of violations.
On March 22, Tencent announced the launch of a new tool called ClawBot, officially integrating OpenClaw into WeChat. Once added, ClawBot will appear in users’ contact lists like a regular contact. Earlier this month, Tencent also introduced QClaw for individual users, Lighthouse for developers, and WorkBuddy for enterprise services.

SP-Life