🎧 News Digest — 2026-07-13 12:00 5 items · ~3 min listen

1. Grok Build CLI sends .env secrets + entire repo to xAI — Wire-level analysis reveals verbatim secret transmission and full-repo upload via two separate channels on a normal consumer login. GitHub Gist · Jul 13
2. Stanford TRACE turns recurrent agent failures into synthetic RL environments — Watches for failure patterns, auto-generates targeted training environments, and trains LoRA adapters on weak spots without disrupting other capabilities. MarkTechPost · Jul 13
3. Prime Intellect releases Verifiers v1 for agentic RL — Composable tasksets, harnesses, and runtimes so teams can mix-and-match evaluation components instead of building one-off setups. MarkTechPost · Jul 13
4. EU pushes for binding social media age restrictions for teens — Ursula von der Leyen: "We need age-appropriate restrictions on platforms." Regulation likely incoming, not just guidance. The Verge · Jul 13
5. Waze gets Gemini-powered conversational reporting and search — Natural language hazard reporting and destination search; promises to be less chatty than typical nav assistants. The Verge · Jul 13

---

🧪 Try This Next

• 🔧 Read the Grok CLI wire-level analysis — understanding what coding agents send to servers should be table stakes
• 🎥 Check out Stanford TRACE on Marktechpost — training agents on failure patterns is directly applicable
• 📝 If you're building agent training pipelines, explore Prime Intellect Verifiers for composable eval