Daily News · 2 min read

Meta AI Updates: May 12, 2026

meta pytorch llama

Quick Hits

  • 1. Santa Clara County sues Meta over $7B in alleged scam-ad revenue: California’s Santa Clara County filed suit in state superior court on May 11, alleging Meta knowingly facilitates and profits from billions of fraudulent advertisements across Facebook, Instagram, and WhatsApp. County Counsel Tony LoPresti claims Meta earns roughly $7 billion per year from scam ads and has tracked some 15 billion fraudulent ad placements, with the AI-driven moderation stack accused of failing to stop coordinated scam networks. The suit seeks restitution, civil penalties, and an injunction. Source

  • 2. Meta Ads AI Connectors open beta goes live globally: Meta’s Ads AI Connectors, including a hosted MCP server and CLI, moved into open beta for all eligible advertisers worldwide, letting third-party AI agents create and manage campaigns through natural language without API integrations. The launch supports ChatGPT and Claude as the initial MCP clients, with more platforms coming, and is a notable concession from a company that historically funneled advertisers into its own automation surfaces. Source

  • 3. PyTorch 2.12 reaches content-finalization milestone, GA targeted May 13: The PyTorch release team confirmed M5 content finalization on May 8 and is on track to ship PyTorch 2.12 GA on May 13, with RC1 binaries already on the test channel. The release introduces an experimental CUDA 13.2 build, deprecates CUDA 12.8, and ships across CPU, ROCm, and XPU targets. Source

  • 4. CUDA 13.2 lands as experimental backend in PyTorch 2.12: Maintainers detailed CUDA 13.2 support as the new experimental build flavor for the 2.12 release, while the long-running CUDA 12.8 line is being deprecated. Downstream libraries pinning specific CUDA wheels will need to update CI matrices ahead of the May 13 GA. Source

  • 5. Class-action consolidation: Llama copyright suits move forward in SDNY: Filings tied to the publisher class action (Elsevier, Cengage, Hachette, Macmillan, McGraw Hill, plus author Scott Turow) progressed in the Southern District of New York alongside the Santa Clara complaint, keeping pressure on Meta over its Llama training data. The plaintiffs lean on court records that allege Meta engineers torrented roughly 82 TB of pirated books and journals, and argue measurable market harm in academic publishing. Source