Radar Daily Briefings
A clearer signal for WordPress and engineering
/\_/\
(=^.^=)
(")_(")
/\_/\
(=^.^=)
(")_(")
Liquid AI releases LFM3 multimodal model family
Liquid AI has released LFM3, an open-weight family of compact multimodal models. It includes text, vision, audio, omni, and decision variants built from the LFM3-1.7B-Base backbone.
The architecture combines gated short convolutions, sliding-window attention, and global attention. Liquid AI also describes hardware-in-the-loop architecture search across phones, Raspberry Pi, laptops, cloud systems, and Qualcomm hardware. The text model supports a 256K context window, while vision, audio, and omni variants add video, speech, and streaming capabilities.
The models are positioned for local agents, specialized fine-tuning, edge inference, and robotics. Reported benchmarks and performance figures come from Liquid AI and are not independently validated in the supplied evidence.
Google DeepMind unveils Genie 4 research preview
Google DeepMind has announced Genie 4 as a research preview for trusted testers. The company describes it as a multimodal world model for real-time interactive simulation, with generated visuals, native audio, and interactive dialogue.
Genie 4 can be steered through natural-language prompts while Gemini models help monitor scenes, break goals into steps, and adapt plans as users intervene. Google says the system supports simulations involving physics, human actions, ambient sound, and multi-step events.
The preview is intended to gather feedback before broader access. Google reports SynthID watermarking for generated video and audio, while acknowledging drift, limited object permanence, and unresolved challenges in complex physics, audio accuracy, text rendering, spatial accuracy, and long-term consistency.
Tailscale investigates SSH privilege-escalation vulnerabilities
Tailscale is investigating security vulnerabilities affecting Tailscale SSH, with most identified issues involving possible privilege escalation. The company says fixes for most findings are implemented, while testing and verification continue.
According to Tailscale, an authorized SSH user may in some circumstances act beyond the permissions of the configured local user, potentially gaining elevated privileges. Exploitation depends on access permissions, system configuration, and other conditions.
Customers are advised to review SSH access policies, restrict sensitive-system access, disable forwarding when unnecessary with TS_SSH_DISABLE_FORWARDING=1, and prepare to update. The full scope, affected configurations, and patched release remain unconfirmed in this update.
Researcher reports six SVG sanitizer bypasses
Gareth Heyes reports bypassing six of eight tested SVG sanitizers in research published by PortSwigger. The article describes JavaScript execution techniques involving SVG uploads and browser parsing behavior.
Reported methods include XML namespace confusion, animation into JavaScript URLs, self-contained XSLT stylesheets, and a Firefox response-header technique. The research also describes Safari importing image files as SVG through the use element, affecting several named libraries.
The findings give engineers concrete cases for testing SVG handling and browser-specific assumptions. The supplied evidence does not establish a confirmed CVE or advisory, and the article says DOMPurify and darylldoyle/svg-sanitizer resisted the described techniques.
OpenAI introduces GPT-6.1 Terra for visual reasoning and computer use
Simon Willison reports that OpenAI introduced GPT-6.1 Terra, a model emphasizing visual reasoning, precise computer use, and clearer writing. The supplied classification identifies the model as released.
The post lists pricing at $1 per million input tokens and $5 per million output tokens, positioning Terra between less expensive and more costly models. OpenAI’s Ted Sanders describes it as a midrange option for API developers, especially for vision, computer use, and communication.
The evidence supports model-selection and cost considerations, but it does not include reproducible benchmarks, architecture details, or detailed evaluation methodology.
WordPress 7.2 Beta 2 is available for testing
WordPress 7.2 Beta 2 is available for download and testing. The release builds on Beta 1 with enhancements, bug fixes, and refinements across administration and developer APIs; the final release is scheduled for December 8, 2026.
Development continues on the Abilities API, including content querying, capability requirements, schema handling, and permissions. WordPress also adds server-side registration capabilities for DataViews and DataForm, plus REST API, admin asset-loading, and SVG icon improvements.
Developers can test through the Beta Tester plugin, WP-CLI, direct download, or WordPress Playground. The beta is not suitable for production or mission-critical sites, and the supplied material does not detail implementation measurements.
WordPress Playground v3.1.61 adds runtime and origin-boundary changes
WordPress Playground v3.1.61 was released on 26 October 2026. The release includes changes spanning browser runtime behavior, origin boundaries, sandbox activation, local development, and testing.
The notes describe SharedWorker runtime use for Personal WP, API calls restricted to the connected window, cross-origin frame messaging, separate internal origins for WordPress frames, and local-folder access checks against the PHP host. The release also enables the WordPress sandbox only with the ?reprint parameter.
The @wp-playground/client package is available at version 3.1.61 through npm. The supplied notes do not include performance data or detailed compatibility guidance.
WordPress invites community review of three Learning Pathways
Learn WordPress is inviting experienced WordPress users, contributors, developers, designers, educators, and community members to review its three existing Learning Pathways: User, Developer, and Designer. The initiative is currently a request for feedback.
Reviewers are asked to identify outdated instructions, software references, tools, workflows, missing concepts, and issues with clarity or structure. Feedback can concern a single lesson and should include the pathway, lesson URL, issue, suggested improvement, and supporting references when available.
Learn WordPress says it will group suggestions, prepare draft improvements, and involve subject-matter experts in checking technical accuracy and priorities. The supplied material does not include completed changes or a timeline.
Wordfence introduces Threat Response Team for active threats
Wordfence has introduced a new Threat Response Team for rapidly emerging threats in the WordPress ecosystem. The company describes it as a small, technically hands-on team focused on customer protection and clear guidance.
People who spot an active threat can report it directly to [email protected]. The announcement does not explain the team’s investigation, triage, or response workflow.
For WordPress site operators and security researchers, the contact path provides a concrete escalation channel. The supplied material does not establish staffing, response times, metrics, or demonstrated outcomes.
WordPress seeks feedback on future release roadmaps
WordPress is asking the community to help shape future release-roadmap posts. The request is open to core committers, extenders, hosting companies, and everyday users, with feedback requested by November 16.
The post says roadmap entries have used roughly the same format and approach for the past three years, with variation based on how mature work areas are. Participants can use the linked survey or comment directly if their feedback does not fit the survey format.
After the deadline, the author plans to summarize the feedback in a separate post or comment. No specific technical change or revised roadmap format has been defined yet.