Sunday, September 6, 2026 · Daily edition
SIEPR’s young-worker signal, FERC large-load orders, and MoChiAgent
Today’s ledger follows a Stanford SIEPR policy brief that reads aggregate AI job loss as still small while new-grad unemployment climbs, six FERC show-cause orders that force RTOs to justify or rewrite large-load interconnection rules for AI halls, Executive Order 14409’s voluntary covered-frontier access window of up to 30 days with an explicit no-licensing line, a Nature Medicine mother–child EHR agent paper with strong retrospective obstetric AUROCs, and UNESCO’s U18 China media-and-information-literacy day after more than 1,500 youth AI projects.
SIEPR: aggregate AI job loss still looks small — sharper signal is young workers
What happened. Stanford Institute for Economic Policy Research published a July 2026 policy brief by Mahoney, McEntarfer, and Wahal — What is really happening to jobs? Separating AI hype from reality — synthesizing recent empirical work for policymakers. On aggregate employment, the authors’ reading is that AI’s current effect is likely small: using IPUMS-CPS and Felten–Raj–Seamans exposure ranks, unemployment in the most AI-exposed quintile rose 0.77 percentage points since 2022, while the least-exposed quintile rose slightly more, by 0.85 points — a broadly soft market, not a clean AI wipeout. They note employment in highly exposed occupations is fairly stable, coding-heavy growth has slowed but stays positive, and online postings for software developers have grown faster than other occupations over the last year. The brief flags a tougher channel for recent graduates: new-grad unemployment reached 5.6% in early 2026, up 1.6 percentage points from three years earlier, and it summarizes evidence that early-career workers in exposed occupations have been hit first while older workers in the same roles held up better.
What to watch. If the aggregate scoreboard still looks muted while new-grad unemployment climbs, the policy print is the youth hiring channel — not another headline jobs apocalypse. This is a SIEPR evidence-synthesis brief, not a U.S. layoff census, not PwC’s job-ad barometer, and not a Federal Reserve FEDS buildout note.
Read the SIEPR policy brief →
FERC’s six show-cause orders: RTOs must justify or rewrite large-load rules
What happened. A White & Case insight alert summarizes six Federal Energy Regulatory Commission show-cause orders issued 18 June 2026 under Federal Power Act Section 206 to all six jurisdictional RTOs and their transmission owners. Following a DOE directive, the orders require the RTOs/TOs to explain why existing open-access tariffs are just and reasonable for interconnecting large loads — defined in the alert as peak load above 50 MW on transmission above 69 kV — or to propose reforms. Five reform buckets are named: efficient study processes; cost transparency and recovery with large-load customers making a minimum financial contribution secured by credit support; dedicated expedited co-location study paths for generation serving electrically proximate loads (within two buses or substations); new transmission services for constrained periods and interim non-firm service while upgrades proceed; and treatment of large flexible loads that can cut demand when the grid is strained. Informational reports on generation adequacy are due in 30 days; full show-cause responses in 60 days; comments on those responses another 30 days.
What to watch. Speed-to-power for AI campuses now collides with tariff design: who studies, who pays, and whether co-located generation gets a faster lane. This is a law-firm summary of FERC large-load show-cause orders, not a named campus gigawatt plan and not a completed tariff rewrite.
Read the White & Case FERC alert →
EO 14409: voluntary covered-frontier access up to 30 days — not licensing
What happened. White House Executive Order 14409, Promoting Advanced Artificial Intelligence Innovation and Security (2 June 2026), sets a national-security frame for advanced AI without creating a licensing gate. Section 3 directs Treasury, NSA (via the Secretary of War), and CISA, with NIST and others, to maintain a classified benchmarking process for advanced cyber capabilities and to let the NSA Director designate a “covered frontier model.” It also orders a voluntary framework under which developers may engage the government on designation; provide government access to covered models for up to 30 days before release to other trusted partners, subject to confidentiality, cybersecurity, insider-risk, and IP protections; and collaborate on trusted-partner early access for critical-infrastructure cybersecurity. The order states explicitly that nothing in the section authorizes mandatory governmental licensing, preclearance, or permitting for development, publication, release, or distribution of new AI models, including frontier models. Section 4 tells DOJ to prioritize 18 U.S.C. 1028 / 1030 / 1343 against AI-enabled unauthorized computer access.
What to watch. The binding U.S. choice on this page is voluntary pre-release access plus criminal enforcement priority — not a model-license regime and not a substitute for already-live EU Article 50 transparency duties.
Read Executive Order 14409 →
MoChiAgent: mother–child EHR agent with strong retrospective obstetric AUROCs
What happened. Nature Medicine published Prediction of maternal and infant outcomes from longitudinal electronic health records with a Mother-Child AI agent (Liu, Zheng, Kang et al.; online 4 September 2026; DOI 10.1038/s41591-026-04694-y). The open-access landing confirms an International Consortium of Digital Twins collaboration and a source-data file comparing MoChiAgent with ChatGPT, Gemini, and OpenEvidence on diagnostic accuracy, evidence traceability, plan completeness, and clinical safety. Independent secondary extract (Maeil Business / MK, 6 Sep) after journal body HTML did not yield full text: core engine MoChiFormer trained on more than 4.4 million longitudinal EHR visits; external validation on about 260,000 maternal and 23,000 infant visits; reported AUROCs for preterm labour 0.91, placental abruption 0.89, and premature rupture of membranes 0.89; paired mother–infant analysis with about 2.8× higher infant jaundice/blood-disease risk when mothers showed hematologic/metabolic risk signs; and LLM + knowledge-search tooling for guideline retrieval from routine labs/EHR rather than costly genetic/imaging tests.
What to watch. Strong retrospective obstetric AUROCs on everyday records still need prospective validation before anyone treats the assistant as bedside proof. DOI/landing locks identity; AUROC and visit counts remain labeled secondary until the full text is read. This is not an RCT and not ECG-CLIP.
Read the Nature Medicine paper →
UNESCO U18 China MIL Day: 41 winners after more than 1,500 youth AI projects
What happened. UNESCO covers a Media and Information Literacy Day at Headquarters in Paris on 4 August 2026 for 41 students from China, including workshops and the award ceremony of the 2026 UNESCO Youth Hackathon U18 China Edition (UNESCO + Yugui / MIL Alliance). The initiative invited under-18s in China to “Play Smart with AI,” drawing more than 1,500 young participants from Shanghai and beyond; the 41 winners were invited to Paris. Sessions covered algorithmic bias, AI-generated inaccuracies, and personalized recommendations, with emphasis on verifying AI-generated information against reliable sources. Projects addressed online fraud, emotional well-being, AI education, and cultural heritage. Page framing: young people as designers of what AI should do, not only users of what it can do.
What to watch. The education signal is verification and design agency for minors — two days before Digital Learning Week and three days before the ICT in Education Prize ceremony. This is a youth MIL event, not the prize ceremony itself and not a student-learning RCT.
Read the UNESCO MIL Day coverage →
Also in today’s ledger
• Keep SIEPR’s 5.6% new-grad unemployment separate from firm surveys and FEDS buildout framing — youth hiring friction can be real while aggregate unemployment still fails the wipeout test.
• FERC’s five reform buckets include two-bus co-location studies, interim non-firm service, and flexible-load options — process design, not metered 2026 data-center kilowatt-hours.
|