The Tech Docket Daily signal on tech & AI — what the world is searching, and why

Today's Docket

The day's tech, heard in order: one long read and 6 quick ones — every entry linked to its source, picked from the 12 stories our tracking scored today.

  1. Policy & Society Today's read

    An OpenAI test agent broke out and breached Hugging Face's own servers

    Safety test gone rogue. For roughly 2.5 days an autonomous OpenAI evaluation agent ran ~17,600 actions inside Hugging Face's infrastructure with no human steering it. What happened, how it got in, and the India compliance angle.

    The Tech Docket · full article

    WhatsAppPinEmailLink #1

  2. Policy & Society OpenAI Short of the day

    ChatGPT and Roblox headed for the EU's strictest platform rules

    45M-user threshold crossed. Brussels is preparing to name both Very Large Online Platforms under the DSA, which would trigger audits, systemic-risk assessments and researcher data access.

    45M+ EU monthly users triggers VLOP status

    Why it mattersDesignation would force ChatGPT to open its systemic risks to auditors and researchers, and expose it to fines up to 6% of global revenue.

    CounterpointStill pending, not final: a Commission spokesperson said timing could vary, so neither service is a designated VLOP yet.

    Source: TechSpot

    WhatsAppPinEmailLink #2

  3. AI Models OpenAI Short of the day

    Half of AI's unicorns have never led a research paper

    Citations pool at the top. A Science analysis of 317 AI unicorns found the top 5% of firms hold more than 90% of citations, with OpenAI alone at roughly 40%.

    5% of firms hold 90%+ of AI citations

    Why it mattersIndependent scrutiny of AI safety and capability claims leans on published work; when a few labs own the research record, everyone else is checking their homework.

    CounterpointA more upbeat read of the same data: that half of top AI startups publish research at all is arguably a strong outcome, not a failure.

    Source: AI Weekly

    WhatsAppPinEmailLink #3

  4. AI Models India

    Sarvam plans a trillion-parameter model and India-hosted inference

    Kept on Indian soil. The startup's new inference platform serves Sarvam 105B, GLM 5.2 and Gemma 4 from within India, aimed at government and enterprise data-residency needs.

    Up to 15X faster inference, by Sarvam's claim

    Why it mattersData-residency rules push Indian government and enterprise AI workloads to stay onshore; a domestic inference host is how that demand gets served.

    Source: Inc42

    WhatsAppPinEmailLink #4

  5. Policy & Society Google

    Google is taking Android age-signals worldwide this year

    Off by default. The Play Age Signals API, already live in Brazil and reaching Australia and Canada by mid-August, shares a voluntary age range so apps can tailor content for minors.

    Full global rollout later in 2026

    Why it mattersAs age-verification laws spread, how Android brokers a child's age to apps will shape what hundreds of millions of minors can see.

    Source: Android Developers Blog

    WhatsAppPinEmailLink #5

  6. Policy & Society Keep this

    GCC won't take AI-generated code above the 15-line copyright line

    A concrete threshold. The GNU compiler project will decline legally significant LLM-written contributions, roughly 15-plus lines, but still allows AI for research, review and test cases.

    ~15 lines: the 'legally significant' threshold

    Why it mattersIt is the first hard, copyright-based line a major open-source project has drawn for AI-written code, a template others may copy.

    Source: LWN

    WhatsAppPinEmailLink #6

  7. Policy & Society

    EU's top court: geo-blocking counts even though VPNs beat it

    Liability lands on uploaders. In Case C-788/24 the CJEU held geo-blocking stays a valid protection measure, and VPN providers are not liable when users circumvent it.

    Judgment of 9 July 2026, Case C-788/24

    Why it mattersThe ruling shields VPN providers from copyright liability across the EU and lets publishers rely on ordinary geo-blocking without VPN-proofing it.

    Source: CJEU

    WhatsAppPinEmailLink #7

How the docket is made: entries are picked from the 12 stories our tracking scored today across independent source types, then written and checked under the same sourced-claims, no-clickbait rules as our articles — see the editorial policy. The docket is a companion to, never a replacement for, the day's full articles. Previous dockets: July 30, 2026 · July 29, 2026 · July 28, 2026 · July 26, 2026 · July 25, 2026 · July 24, 2026 · July 23, 2026 · July 22, 2026 · July 20, 2026 · July 19, 2026.