AI news: Grok Voice 3.0, Claude Code fixes and Deezer’s AI-music filter — 15 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

Grok Voice 3.0 is the clearest new audio-model release in this scan; Claude Code’s update and Deezer’s forthcoming AI-music filter affect different parts of the creator workflow. This edition covers 12 October at 13:26 through 15 October at 12:14 Europe/Istanbul (UTC+3). The source dates are 13–14 October; future rollout dates remain future dates.

Grok Voice 3.0: API availability and an automatic alias upgrade

SpaceXSI’s 14 October announcement makes Grok Voice 3.0 available through the Voice Agent API at $0.08 per minute, unchanged from its predecessor. The provider operates the x.ai API and says existing grok-voice-latest sessions have moved to the new model. Release announcement.

The dated API release notes confirm that grok-voice-latest now points to grok-voice-3.0, with the API interface unchanged. They also deprecate grok-voice-2.0 and direct developers to the new model or latest alias. Deprecation is not, by itself, an established shutdown date.

For voice applications, the concrete change is the model behind an existing alias. Check the selected model and retest turn-taking and tool behavior before assuming unchanged application results. The vendor’s comparative quality and latency claims are not MUVO benchmarks. The announcement does not establish identical country or plan access across consumer Grok and API products.

Claude Code 2.1.299: MCP policy and session reliability fixes

Anthropic published Claude Code 2.1.299 on 14 October at 02:49 UTC. Its release notes include fixes for MCP policy checks, permission-rule matching and Remote Control delivery, alongside terminal idle compaction. Official release.

The notes say blocked MCP servers are now stopped before connection in several affected paths. They also narrow MCP allow rules to the named server, and change prompt-submit hooks to run before a skill or custom slash command expands. These are documented behavior changes for developers and administrators to review against their own configuration, not evidence of an active compromise.

Other fixes address Remote Control prompt delivery and memory use after resuming agent-heavy sessions. Terminal idle compaction targets 200,000-plus-token conversations after roughly 54 idle minutes. These are vendor release claims; MUVO has not tested them.

Deezer: a forthcoming filter for fully AI-generated music

Deezer announced on 13 October that listeners will be able to exclude music identified as fully AI-generated from search results, with worldwide rollout planned before the end of 2026. The toggle will cover identified tracks, albums and associated artists in the affected results, and can be switched off again. Official announcement and FAQ, in French.

This adds search filtering to Deezer’s existing recommendation and editorial-playlist exclusions. It does not announce universal catalog removal or establish how every partially AI-assisted recording will be classified.

Analysis: Catalog availability and discovery are different. Creators should distinguish inclusion in a service from visibility through recommendations or filtered search. The future audience impact is not established.

Google: Gemini-powered message protections have staged access

Google’s 14 October Android announcement adds an Enhanced protection setting for Google Messages, with rollout in the coming weeks. It describes using cloud Gemini through Private AI Compute for suspicious conversations with unknown senders, plus on-device Gemini Nano for unfamiliar-link checks. Official Android announcement.

Eligibility differs by feature. The cloud-analysis footnote lists Pixel 6+, Galaxy S24+, Flip/Fold6+ and other AICore devices; the Gemini Nano link-check footnote lists Pixel 10+, Galaxy S26, Flip/Fold8 and selected devices. Check regional availability separately.

Chrome/WebView screen-share protection is scheduled for Android 15+ in October; call-forwarding protection follows on selected devices in early December. These are separate rollout dates.

Claude: US voting-information integration is forthcoming

Anthropic’s 14 October midterms update says Claude.ai will begin drawing voting information from Democracy Works for US users later in October. It describes state-specific information such as polling-location and identification requirements, alongside the existing TurboVote banner. Anthropic’s announcement.

The company also plans to display current results on election night and afterward. These are announced consumer-app integrations; they are not evidence of an equivalent public API feature or completed rollout at this cutoff. The practical distinction is access to named source information in answers, not a guarantee that every generated political response is correct. The announcement’s usage analysis is the company’s own research, not a survey conducted by MUVO.

ElevenLabs: new guidance explains audio-detection limits

ElevenLabs’ 14 October guide distinguishes statistical speech classification from checking a provider’s watermark. It says its Audio Detector checks ElevenLabs watermarks and does not identify those of other voice providers; third-party synthesizers offered through ElevenLabs generally are not watermarked. Audio-detection guide.

The guide separately describes its public AI Speech Classifier as free without sign-in, assessing the first minute for likely ElevenLabs generation. It warns that this classifier cannot reliably classify Eleven v3 audio and is less accurate for heavily compressed or edited speech.

These limitations matter when checking synthetic voice content: failure to detect one provider’s watermark is not proof that a recording is authentic. This is newly published guidance, not a verified new detector-model launch. MUVO has not tested its accuracy or confirmed enterprise access terms.

OpenAI and Amp: voice control of a third-party agent workflow

OpenAI’s 14 October Amp story describes Calling Puck using GPT-Live-1 for real-time voice interaction with an assistant that launches and manages agents. The account also discusses Amp’s Sign in with ChatGPT integration. Original customer story.

The operator is Amp, not a newly launched OpenAI consumer app. The story says Amp has been testing Astra Ultrafast; it does not establish general availability of that mode inside Amp. Nor does it specify complete plans, country eligibility or an exact original launch date for each integration. Treat the story as a newly published workflow account, not an announcement that all ChatGPT subscribers now receive the same third-party features.

ElevenLabs: support-agent case study separates tools from escalation

ElevenLabs’ 14 October account of its own support system describes agents using account tools, structured procedures and human escalation. It also describes simulation tests and a period of observing proposed actions before allowing production actions. Company workflow account.

The useful developer detail is how an agent is bounded and checked: identifying an account, obtaining relevant information, following a procedure and escalating exceptions are separate steps. The company’s support-volume and satisfaction figures are self-reported. This article does not verify individual customers’ outcomes or announce a new universal refund policy, service commitment or feature entitlement.

Sources

• Grok: release announcement and API release notes.
• Anthropic: Claude Code 2.1.299 and US election-information plans.
• Deezer: AI-music filter announcement.
• Google: Android scam protections and eligibility footnotes.
• ElevenLabs: detection guidance and support-agent workflow.
• OpenAI: Amp customer story.

The scan included music, video and remix tools. Older Suno Studio and Runway releases are not new stories merely because their pages were retrieved again. Some specialist sources were inaccessible or lacked a dated announcement; those gaps remain in the research ledger. For earlier verified developments, see 14 October’s news edition.

Vendor performance claims, detection claims and future rollout commitments remain attributed to their providers. Service disruptions and customer reports are covered separately in today’s AI incidents.

Update timeline

• 15 October 2026, 12:14 Istanbul: Research cutoff for this edition. Eight qualified developments consolidated. Publication followed later on 15 October.