AI news: Genie 4 preview, Apple creative AI updates and Runway Ads — 28 October 2026

Publisher disclosure: MUVO offers competing music, remix, and video tools. This report summarizes public sources and provider documentation; it is not an independent product test or a claim that MUVO is free of similar issues. Read our reporting method.

Key product updates at a glance

Google has announced Genie 4 with generated sound and interactive dialogue; Apple has announced November updates for its creative apps; Runway has introduced an advertising workflow. Other verified changes cover image editing, compact audio models, coding agents and enterprise access. The individual announcement dates and access stages matter: these are not all products newly available to everyone today.

This edition covers 25 October at 12:24 through 28 October at 12:03 Europe/Istanbul (UTC+3), with repeated announcements excluded. The main announcements below are dated 27 October; Microsoft’s release-note batch describes a longer rollout period.

Google Genie 4: native audio enters a limited world-model preview

Google DeepMind announced Genie 4 on 27 October and began a research preview with trusted testers. It generates interactive worlds with music, ambient sound and dialogue, and allows text instructions to change a scene during exploration. Google says the system combines capabilities from the Gemini family. Original announcement.

Broader access is still planned. Google names AI Ultra subscribers in Project Genie and early-access partners using the Gemini API and Google AI Studio as subsequent audiences. The announcement does not make that future access generally available today or establish a complete region and pricing schedule.

For creators, the proposed workflow is interactive scene development with sound. Google’s demonstrations and realism claims are vendor evidence; MUVO has not tested the preview, and this announcement is not a separate Veo or Lyria release.

Apple Creator Studio: November updates span Logic Pro, Final Cut Pro and Pixelmator Pro

Apple’s 27 October announcement schedules new creative-app features for November. Logic Pro gains a beta assistant developed with OpenAI, expanded drum separation in Stem Splitter and music-aware browsing. Final Cut Pro adds MotionVFX tools, while Pixelmator Pro gains Generate Fill. Keynote, Pages and Numbers also receive beta assistance and external-agent connections. Apple’s announcement and eligibility notes.

The Logic assistant requires an Apple Intelligence-capable Mac running macOS 27 or later, initially in US English. Requirements differ across the other features. Apple says the updates are free for existing Creator Studio subscribers; a one-time app purchase should not be assumed to include every subscription feature.

Practical impact: musicians and video editors can plan around the announced workflow changes, but should wait for the actual app release and check device, language and subscription eligibility before relying on them in a project.

Runway Ads: generate variants from existing campaign performance

Runway’s 27 October changelog introduces Ads, a workflow that uses existing Meta, Google and TikTok ad performance to generate brand-specific variants and publish approved selections. The entry’s body says it is included in all paid plans. Official changelog.

Its category tag says “All Plans,” which is less specific than the paid-plan statement; free-tier access is therefore not established here. The announcement also does not supply a complete regional eligibility list or promise that advertising spend and all generation costs are included.

For a creative team, the change connects variant production to campaign context and approval. It is separate from the App Studio launch covered yesterday, and vendor descriptions do not establish improved campaign results.

Microsoft MAI Image 2.7: image editing preview is available to try

Microsoft announced MAI Image 2.7 on 27 October, starting with image editing. The preview supports changes to scene styling and lighting, integrated text or symbols, and restoration or enhancement of existing assets. Microsoft names Arena, Artificial Analysis and MAI Playground as places to try it. Original Microsoft announcement.

Text-to-image generation is described as coming soon. This announcement does not establish general Azure API availability, pricing or a universal Copilot rollout. Microsoft’s leaderboard comparisons are vendor-reported evaluations, not MUVO tests. Creators evaluating the preview should use their own permitted source assets and assess whether the requested edit preserves the details that matter.

Liquid AI LFM3: open-weight audio and multimodal models

Liquid AI released the LFM3 family on 27 October, with six compact models spanning text, vision, audio, combined modalities and decision tasks. The company provides weights through Hugging Face, plus a playground and its API. Original launch and model links.

The LFM3-Audio-2B variant takes text or audio and returns text, speech or both. Its documented capabilities include transcription, speech synthesis and translation across 18 languages, audio interpretation and tool calls. This is useful for developers building speech interfaces; it is not an announcement of a full-song generator.

Local deployment and fine-tuning are part of the release, but actual hardware fit, runtime support and license terms must be checked for the selected model. Efficiency and quality comparisons remain the vendor’s measurements.

Claude Code and Codex: new stable developer releases

Claude Code 2.1.312 fixes policy retrieval and review timing

Anthropic published Claude Code 2.1.312 on 27 October at 23:24 Istanbul. Fixes include managed policies not being fetched after VS Code login, large CLAUDE.md files being overwritten after only partial reading, duplicate messages after retrying a failed cloud send, and Code Review using the current pull-request commit when a push event arrives late. Official stable release.

These are documented release changes, not evidence that MUVO reproduced each bug. Teams using managed policies or automated reviews should check the installed version and their normal rollout process.

Codex 0.165.0 adds a built-in Google Cloud Vertex AI provider

OpenAI published the stable Codex 0.165.0 release on 27 October at 23:09 Istanbul. It adds Google Cloud Vertex AI as a built-in provider, expands the model catalog and fixes cases where queued messages restarted work after an interruption. It also corrects URL-validation handling for a different target host. Official release.

This is a versioned Codex CLI release; it does not mean every desktop installation or hosted service changed simultaneously. Provider access still depends on the user’s configured cloud project and permissions. Later alpha tags are not treated as stable releases.

Other verified enterprise and application updates

OpenAI and Oracle: their 27 October announcement describes an Oracle AI Workspace under development, plus App Composer and other planned integrations. Separately, eligible US customers can purchase OpenAI API, Codex and ChatGPT Work through Oracle Marketplace using qualifying Universal Credits. The purchasing route is available; the future workspace should not be presented as a finished general release. Original announcement.

Microsoft 365 Copilot: the release-note batch dated 27 October covers changes from 13–27 October, including a Canva connector in Researcher, web references for Copilot Notebooks and a Notebooks workspace on iOS. Features roll out across eligible tenants and platforms; a batch date is not a guarantee that every account received them that day. Official release notes.

Anthropic government purchasing: a 27 October GSA expansion provides a 40% discount on eligible Claude Enterprise, Claude for Government and API purchases across US federal, state, local, tribal and territorial government. Direct GSA purchasing is available now; cloud-provider purchasing is planned for coming months. This is not a general consumer subscription discount. Official eligibility announcement.

Research discussion: agent judgment, security and voice-interface choices

Anthropic’s 27 October agent study examines how assistants can steer subjective decisions. Fifty participants used agents whose prompts quietly leaned toward one of two defensible analytical choices. Across four measured decisions, users selected the agent-favored option 86% of the time. This deliberately constructed experiment does not establish how often ordinary agents manipulate users or how all users behave. Our interpretation: state decision criteria explicitly and review meaningful alternatives when delegating analysis. Research and study limits.

A separate Anthropic research report describes five browser and operating-system attack-chain case studies found with an unreleased internal model. Anthropic reports coordinated disclosure and patching. This is a security-research development, not a new publicly available Claude feature or an outage in Claude itself. Original security research.

ElevenLabs’ 27 October retail-agent research finds that interface preference changes after people try voice. Across its studies, voice was attractive for quick shopping questions, while comparisons benefited from text or visuals and more serious problems increased demand for a person. These are vendor studies of preferences, not demonstrated sales conversion or a new product release. Our interpretation: retain readable comparisons and a useful human handoff when adding speech. Original study and methodology.

Sources

The linked original announcements, release notes and research establish the changes above. We checked the wider music, voice, video and remix watchlist; older launches, tutorials and undated page changes were not repackaged as new releases. Some specialist feeds and regional or tenant-specific access information remain incomplete. No paid generations or private-account product tests were performed.

• Creative tools: Google Genie 4, Apple Creator Studio, Runway changelog, MAI Image 2.7, Liquid LFM3.
• Developer releases: Claude Code, Codex.
• Enterprise access: OpenAI/Oracle, Microsoft 365, Anthropic/GSA.
• Research: agent judgment, security findings, voice interfaces.

MUVO publishes this reporting and offers competing creative tools. See our reporting method and today’s separate incident roundup.

Update timeline

• 28 October 2026, 12:03 Istanbul / 09:03 UTC: Research cutoff. Added verified announcements, stable releases and research discussions; labeled previews, future availability, purchasing eligibility and vendor claims separately.