All Updates
Page 1841 of 1901
February 19, 2026
RAM Crunch May Kill Products, Companies
Phison CEO Pua Khein-Seng warns that a severe RAM shortage could force companies to cut product lines in H2 2026 and even cause some to fail due to component unavailability. The statement came in a Chinese TV interview on Next TV, confirmed by sources amid machine-translated headlines. Phison produces controller chips for SSDs and flash memory devices.
Qwen-Code v0.10.4 Nightly Released
Qwen-Code announces the release of v0.10.4-nightly.20260219.c6a723ef on GitHub. The full changelog is available comparing from v0.10.5 to this nightly build. This update targets developers working with coding-focused LLMs.
Microsoft Splits Win11 Canary into Dual Tracks
Microsoft split the Windows 11 Canary channel into 28000 series for 26H2 feature previews and new 29500 series for platform development. The 29500 track targets kernel, drivers, and hardware compatibility with high instability risks. Upgrading is one-way, requiring clean install to revert.
GitLab Security Dashboard Adds Remediation Trends
GitLab 18.9 updates the Security Dashboard with trend tracking, vulnerability age distribution, and risk scoring. It helps prioritize remediation using factors like EPSS and KEV scores. Teams can monitor velocity, identify training needs, and reduce manual reporting.
GitLab Exposes North Korean Dev Scams
GitLab's Threat Intelligence Team published a report detailing North Korean actors' use of the platform for Contagious Interview malware campaigns and IT worker fraud since 2022. They banned offending accounts in 2025 and shared insights into malware repos, synthetic identities, and financial records. No customer action required; GitLab remains secure.
GitLab Duo Self-Hosted and BYOM Launch
GitLab 18.9 enables self-hosted Duo Agent Platform for online cloud licenses with usage-based GitLab Credits billing. It supports data residency and control for regulated industries. Bring Your Own Model (BYOM) integrates custom models via GitLab AI Gateway for full enterprise governance.
CDLM: 14x Faster Diffusion LM Inference
Together AI's Consistency Diffusion Language Models (CDLM) overcome standard diffusion LM limitations by enabling exact block-wise KV caching and trajectory-consistent step reduction. This post-training recipe delivers up to 14.5x latency improvements without quality loss. It makes diffusion LMs practical for real-world use.
February 18, 2026
Sony Monetizes PS5 Users to Dodge Memory Hikes
Sony plans to offset AI-driven memory price surges by monetizing 92M existing PS5 users via software/network services. Hardware prices stay stable, but PS Plus hikes likely. PS6 delayed to 2028-2029 awaiting market normalization.
AWS Bedrock AgentCore Enables Unified Intelligence
AWS Machine Learning Blog post demonstrates building unified intelligence systems with Amazon Bedrock AgentCore. It showcases a real-world implementation called Customer Agent and Knowledge Engine (CAKE). This helps integrate agents and knowledge bases effectively.
Microsoft AI Inserts Typos in Old Diagram
A programmer's 15-year-old diagram ended up on Microsoft Learn with embarrassing typos apparently added by AI. Vincent Driessen highlighted the curious errors generated during processing. This reveals potential pitfalls in AI content handling.
Palantir Partial Win vs Ex-Workers' AI Startup
A judge sided with Palantir, ruling three ex-employees likely violated confidentiality and non-solicitation agreements by founding AI startup Percepta. However, the court declined to halt their operations at the new company.
Meta AI Smartwatch Launch Planned for 2025
Meta plans to launch an AI-enabled smartwatch with health tracking later this year, reviving a canceled project codenamed Malibu 2. It precedes delayed Phoenix MR glasses pushed to 2027. Faces competition from Apple, Google, and others in wearables.
KnowBe4's Deepfake Boss Videos for Phishing Training
KnowBe4 launched a Japanese version of training using deepfake videos and audio of its executives to simulate sophisticated impersonation attacks. The program allows users to experience advanced AI scams hands-on. It aims to sharpen judgment skills against AI-driven fraud.
Japan IBM's AI Strategy: 3 Pillars for Legacy Modernization
Japan IBM announced its AI strategy, positioning 2026 as the inaugural year for full AI integration into system development and business processes. The company is launching an AI platform equipped with development support tools and governance features. It emphasizes strengths in hybrid environments for large enterprises.
Meta Plans AI Smartwatch Launch This Year
Meta is reportedly set to launch the Malibu 2 smartwatch with health tracking and AI features later this year, ahead of updated Meta Ray-Ban smart glasses. The company has delayed its Phoenix mixed reality glasses until 2027 to streamline its AR/MR roadmap. This follows the cancellation of prior smartwatch plans in 2022 due to technical and cost issues.
48-Hour Mandate for Removing Abusive Images
The government proposes a new law requiring tech firms to remove abusive images within 48 hours. Intimate image abuse will be treated as a more severe offense. This aims to strengthen protections against image-based sexual abuse.
Microsoft Flags AI Memory Poisoning Attacks
Microsoft warns of 'AI Recommendation Poisoning,' a technique abusing generative AI memory functions via malicious URLs to bias recommendations toward specific companies. Over 50 cases have been confirmed, posing a new threat to AI neutrality.
WD Invests $74M in Thailand HAMR R&D
Western Digital approved for 23 billion THB (~$73.8M) investment in Thailand HAMR HDD R&D center. The project expands existing facilities to advance heat-assisted magnetic recording tech. Aims to boost local HDD production capabilities and value.
TPG CEO on AI Opportunities Amid IPO Uncertainty
TPG CEO Jon Winkelried discussed opportunities in AI, worries around software investments, and uncertainty in IPOs during a Bloomberg Deals interview. TPG manages over $300 billion across diverse strategies.
GEA Matches Human AI Agents at Zero Cost
UC Santa Barbara researchers developed Group-Evolving Agents (GEA), enabling AI agent groups to evolve collectively by sharing experiences and innovations. GEA outperformed existing self-improving frameworks on coding and software tasks, matching or exceeding human-engineered systems. It eliminates inference costs and overcomes individual-centric evolution silos.