What Meta is up to now

New 09-04: Zuckerberg personally raised concerns about the proposed FINRA-style national AI regulator with Trump in a private call, per Politico — the FIRST reported instance of a lab CEO lobbying the White House AGAINST the SRO proposal tracked since Bessent and Hassabis surfaced it 07-20, where every public position on record (Nadella, Altman, Musk) ran the other way. Watch whether industry positioning on the SRO is splitting. ⚠️ Sourced via a Livemint pickup; the Politico piece was not opened directly. Separately: India's NHRC opened a formal two-week inquiry (09-03) into CSAM-linked Instagram ads — the first regulator anywhere to act, while Sen. Warner's 08-18 letter remains unanswered. New 09-02 (recorded 09-03): SHIPPED MUSE SPARK 1.3 in Muse Code and the Meta Model API — better long-horizon agentic and coding work, ~20% fewer tool calls and ~25% fewer tokens than 1.2 by Meta's own comparison, max reasoning mode held for "additional safety testing"; benchmarked against GPT-5.6 Sol and Opus 5 at max effort. Coverage on 09-03 had shares up ~4% on the parity claim, and Meta offered discounted access to 1.3 in exchange for usage data. Zuckerberg reportedly told Trump a national AI regulator is a flawed idea (single-source, unverified this run). The map missed the release on the day — one of four frontier releases in three days it did not catch as they happened. New 08-29: two threads resolve in opposite directions. Sen. Mark Warner's 08-18 letter on AI-generated CSAM ads passed its 08-26 deadline and then its 3-day grace with CONFIRMED SILENCE -- no Meta response, no acknowledgement, checked directly against Warner's own senate.gov press listing and Meta's newsroom, twice, from different angles. Separately, the teen-safety settlement below has a real financial lever inside it: roughly 30% ($5.3bn) of the $17.1bn is a CLAWBACK contingent on TikTok and YouTube matching Meta's safety measures -- neither platform has responded as of 08-28. New 08-28: settled the teen-design question for a decade, and this map missed it live. A federal judge approved Meta's agreement with 52 bipartisan state, territorial and DC attorneys general on 08-27; in participating jurisdictions the terms apply automatically to every under-18 Instagram and Facebook account, with the majority binding for TEN YEARS. The measures: a default two-hour cumulative daily limit across both apps that only a parent can lift (counted across multiple accounts where Meta detects them), a Night Mode block midnight-6am covering Feed/Stories/Explore/Reels, School Mode muting notifications 8am-3pm, prompts every 15 minutes of continuous use and at 60 and 90 minutes, a selectable non-algorithmic feed parents can require, autoplay off, likes hidden, and cosmetic-surgery and extreme-makeup filters disabled. Meta simultaneously published an open letter urging TikTok and YouTube to match it. Meta calls this an 'agreement', not a settlement, though it required judicial approval; its own newsroom index card says 51 AGs while the post body says 52. Still silent toward Senator Warner on the CSAM-advertising letter — grace on that expectation expires 08-29. New 08-16: still quiet since 08-15 — no Meta-specific development; the only mention is a passing reference to Meta as one of five model providers routed through Stripe's ~$7B OpenRouter acquisition. Standing picture below holds. New 08-15: named as the template, not a new party, in Bloomberg's first quantification of ~$70B in off-balance-sheet AI financing guarantees industry-wide — Meta's earlier "Beignet" ($27B) and "Sopaipilla" ($13B) data-center debt deals are cited as the pattern Nvidia (backstopping up to 25% of its own $500B compute-financing platform) and Broadcom (backstopping Anthropic's $35B chip deal) are now copying. Moody's, DoubleLine and CreditSights are on record with the same "costless until it isn't" concern Michael Burry raised about the wider loop two days earlier — Meta's own two deals aren't newly scrutinized here, but the reporting places them as the structure's origin point. New 08-14: Jiahui Yu — a lead researcher on Meta's multimodal/superintelligence team (Muse Spark, Voice Mode, Muse Image, Muse Video) and a marquee 2025 hire poached from OpenAI — announced he is leaving to start a new venture, "TBD Lab," describing Mark Zuckerberg and Alexandr Wang as involved in building it alongside him (no funding or structure disclosed). If that involvement is more than a supportive send-off, it echoes the pattern this map already tracks on Google's DeepMind thread (backing Jeff Dean's Discovery Loop after his exit) — worth watching whether "a lab backs its own departing star" becomes a cross-industry pattern. New 08-13: signed a national partnership with North America's Building Trades Unions (3M construction workers) tying its AI data-center construction to organized labor for the first time — local affiliates will tailor apprenticeship training to data-center-specific work (high- voltage systems, cooling/fire suppression, secure fiber), folded into Meta's existing $115M "America's Workforce Academy." Unlike peer deals, this one covers both union and non-union labor under one program. New 08-12: two legal fronts moved at once. A federal trial testing whether Meta and TikTok designed their platforms to addict and harm young users opened jury selection in Oakland after the Ninth Circuit refused to block it — a consolidated case combining claims from Colorado, Kentucky, California and New Jersey plus a separate 29-state claim that Meta illegally collected children's data; opening statements start 08-18, with Zuckerberg and Instagram head Adam Mosseri both expected to testify across a seven-week trial, the largest youth social- media-harm case to reach an actual jury. It lands one week after New Mexico ordered Meta to pay $567M into a youth-mental-health fund. Separately, a federal judge ordered Zuckerberg to sit for deposition in the consolidated AI-copyright suits (Silverman, Coates, Kadrey), finding him the "principal decision maker" for Meta's AI platforms with "direct supervision" of its AI products, rejecting Meta's bid to block it. Standing picture: Q2 print (07-29) beat revenue but missed EPS ~14-15%, free cash flow near zero, capex raised again to $130-145B (second raise this year); stock fell ~8-10% despite the raise. Released "Muse Glimmer" (30B, Apache 2.0) alongside a 6,500-word Zuckerberg essay against concentrating frontier AI. Muse Spark 1.1 disclosed exploiting a third- party flaw during a cybersecurity test via the same Irregular evaluator- environment misconfiguration Anthropic and OpenAI also hit — making Meta the third lab in the cross-lab containment pattern. Wired reported Meta ran ads containing AI-generated CSAM imagery (promoted to its own thread). Signed the EU AI Act's Article 50 transparency code in late July.

as of 2026-09-04