New Mexico court orders Meta to pay $567M and build child safeguards — including AI chatbot limits.
What happened: In the second phase of New Mexico’s landmark child-harm case, Judge Bryan Biedscheid ordered Meta to pay $567 million — mostly for youth treatment services — on top of $375 million in civil penalties jurors already imposed in March. The order also requires product changes for New Mexico users under 18: a 90-hour monthly time limit, AI chatbot restrictions, mandatory warnings, stronger age-assurance tools, default-hidden like counts for minors, and tighter barriers between children’s accounts and unconnected adults. Meta says it will appeal; the company argues it already has teen safety tools and that the claims misrepresent its record.
Why it matters: Courts are writing enforceable product specs where statutes lag. The measurable record is whether age assurance and chatbot limits actually ship, whether other states copy the remedies, and whether Meta’s appeal freezes the redesign.
Rippling’s AI bill hit 40% of R&D payroll — so it productized token ROI tracking.
What happened: After encouraging company-wide “tokenmaxxing,” Rippling found AI spend growing ~80% month-over-month and on track to equal 40% of R&D headcount compensation — with 10–15% of employees driving ~60% of spend and one engineer at $50,000 a month. It negotiated vendor caps, built an internal AI gateway to route work off default frontier models, and launched AI Spend Console to map token use to employees, teams, PRs, and rework. Internal usage later returned near peak volume (~600B tokens/month) while July’s token cost fell to about 37% of April’s; spend as a share of headcount budget dropped toward 10–15%.
Why it matters: AI labor economics is shifting from headcount cuts to unit-cost control. The measurable record is whether companies gate access on productivity metrics, how aggressively they route to cheaper models, and whether non-engineering roles keep broad AI tools once ROI must be proven.
AI data-center power is a regional U.S. grid problem — not proof of a global electricity takeover.
What happened: A CleanTechnica analysis separates two claims often bundled together: AI is already a material new electricity load in concentrated U.S. clusters, and long-range 2030 forecasts should be treated as planning scenarios rather than settled facts. It cites IEA figures of roughly 415 TWh of global data-center electricity in 2024 (~1.5% of world use) rising toward about 945 TWh by 2030 (still under ~3%), while noting Berkeley Lab-style U.S. pathways in which data centers could approach double-digit shares of national electricity under central cases. The hard constraint is local: substations, transformers, transmission corridors, and who pays if announced load never fully materializes.
Why it matters: Panic and dismissal both fail the denominator test. The measurable record is interconnection queues, cost allocation rules, project attrition, and whether efficiency gains bend demand before utilities socialize speculative wires.
Flock makes ALPR audits and case codes mandatory, cuts default retention to 7 days — ACLU still unconvinced.
What happened: Facing reports that officers used its AI license-plate network to stalk people, Flock Safety is requiring previously optional Audit Assistance (flag abnormal searches and lock accounts pending review), requiring case codes for searches, and cutting default data retention from 30 days to seven, with an “Evidence Mode” exception for case-tied data. CEO Garrett Langley told The Verge the company “got this one wrong” by waiting on lawmakers. The ACLU says the package still looks more like PR than structural reform, warning Evidence Mode could swallow the retention cut and that audit tools need independent evaluation. Flock operates on the order of 120,000 cameras; dozens of communities have already paused or canceled contracts.
Why it matters: Networked vehicle AI is becoming a de facto national surveillance layer. The measurable record is retention defaults in the field, independent audit efficacy, contract cancellations, and whether state ALPR bills harden beyond vendor promises.
Nature: “AI scientists” still fail when the original authors grade the work.
What happened: A Nature news feature covers a late-July arXiv study led with Princeton’s Sayash Kapoor that stress-tests automated research systems with “shadow evaluation.” An agent built around Claude Opus 4.8 got six days and $3,000 in compute per task to pursue research questions from two NeurIPS submissions, then the original human authors scored the outputs. Engineering endurance looked real — hundreds of experiments, literature review, some self-caught hallucinations — but scientific quality did not: overall marks were 2/6 and 1/6. Common failure modes included settling early on weak hypotheses and self-review that was not harsh enough to force a pivot.
Why it matters: Peer-review theater is a low bar for claims that AI can automate discovery. The measurable record is author-judged novelty, error under distribution shift, and whether labs stop equating workshop acceptances with breakthrough capacity.
Duke’s August syllabus guidance: every course needs an explicit generative-AI policy.
What happened: Duke’s Center for Teaching & Learning refreshed AI syllabus guidance (updated August 9) tied to the 2026/2027 syllabus template from the Office of Undergraduate Education. Under the Duke Community Standard, unauthorized generative-AI use is academic dishonesty — but “unauthorized” is course-defined. Instructors are told to publish a clear stance, explain the rationale, support AI literacy (hallucinations, citation, prompting), and choose along a continuum from full prohibition to free use with or without acknowledgement. The guidance explicitly discourages relying on AI-detection software alone because of false positives and bias.
Why it matters: Campus AI governance is moving from blanket bans to assignment-level contracts. The measurable record is how many syllabi ship concrete policies, whether detection tools remain secondary evidence, and whether students get literacy instruction rather than only penalties.
