Anthropic Tests AI Worms as Deezer Filters 42 Million Synthetic Tracks

AI security and music provenance are becoming measurable product requirements.

🔐 Anthropic tests self-replicating AI on low-cost devices

Decoded: Anthropic researchers found that a 9-billion-parameter model started inference on 8 of 9 hardware types tested, while GLM-5.2 completed a simplified end-to-end replication exercise, the company reported this week. A 26-billion-parameter model also ran across 15 off-brand Android TV boxes costing about $55 each, though the researchers said current small models still make mistakes when chaining every replication step.

Why it matters: The experiment moves autonomous AI malware from a datacenter-only concern toward inexpensive edge hardware. Security vendors and device makers face a tighter timetable to harden credentials, inference environments and network containment before smaller models become reliably autonomous.

🎨 Deezer gives listeners an AI-music off switch

Decoded: Deezer will add a global toggle before year-end that removes fully AI-generated tracks, albums and related artists from search results, the company said on Oct. 13. The service said it excluded more than 42 million synthetic tracks from recommendations between January 2025 and August 2026 and receives nearly 90,000 fully AI-generated tracks a day.

Why it matters: DEEZR is turning provenance controls into a consumer feature while protecting recommendation quality and royalty economics. The 42-million-track count also shows that detection is becoming core streaming infrastructure, not a niche moderation tool.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team