DOE draft transmission study names data centers as a first-order reason the U.S. needs more wires.
What happened: The U.S. Department of Energy’s Office of Electricity released a draft 2026 National Transmission Needs Study for a 60-day public comment period ending September 7, 2026. The draft says there is a pressing need for additional electric transmission because of load growth from data centers, expanding domestic manufacturing, large industrial loads, and electrification. It finds that the majority of transmission congestion is concentrated in about 5% of hours — especially during large day-ahead to real-time price swings, high net load, cold weather, and high intermittent generation. Within-region congestion-relief potential is highest in NYISO, NorthernGrid South, and MISO. Interregional value shows up on links such as WestConnect–SPP, NorthernGrid–WestConnect, and ISO-NE–NYISO. The study notes that MISO, SPP, PJM, and ERCOT have recently approved some of their largest transmission portfolios, with MISO’s 2024 portfolio described as the largest ever in the U.S. The assessment draws on more than 120 recently published reports and is not meant to name specific line solutions or displace existing planning processes.
Why it matters: AI-era electricity demand is now written into the federal transmission ledger as a planning driver, not a side note. The measurable record is comment-window outcomes, who pays for wires, and whether congestion relief reaches the regions where large loads are landing first.
Hollywood’s new AI studios pitch cheaper films — and workers hear displacement.
What happened: On a Culver City set for the AI-enabled horror film Touch the Grass, the new studio Promise used human actor Tori Thomas on a small stage while Chinese model Seedance 2.5 generated live backgrounds and effects. Promise is backed by Google, Silicon Valley venture capital, and Disney. CEO George Strompolos estimated hybrid AI films could cost 20% to 50% less than conventional production. Netflix said last week it had used AI in 300 of its 1,000 titles in 2026. Ron Howard’s Imagine has an AI studio, Obsidian, aiming for roughly 30% to 40% cost savings on an AI-enabled animated POW documentary. Brad Pitt said AI could help get mid-budget $40m–$90m films made. Directors including Christopher Nolan and Guillermo del Toro have criticized generative film tools; Rep. Laura Friedman warned last month not to wait until “tens of thousands of American workers” are displaced. Oscar-winning effects veteran Joel Hynek called the shift as big as optical-to-digital. AI-native filmmaker Kavan Cardoza predicted AI actors could match real ones next year — a claim SAG-AFTRA is expected to contest.
Why it matters: Film and TV are becoming a live labor market for generative tools, not a demo reel. The measurable record is budget cuts that stick, residual and consent rules, and whether “micro studios” create new creative jobs or hollow out mid-budget crews.
Virginia orders Dominion to put more data-center wire costs on the loads that caused them.
What happened: In a July 31, 2026 order covered this week, the Virginia State Corporation Commission told Dominion Energy to change transmission cost allocation so certain “direct connect” substations and transmission lines needed by new large-load facilities — including data centers — are paid through a mandatory contribution in aid of construction (CIAC). Hyperscalers including Google and Amazon had asked for voluntary payments; the SCC made them mandatory. In the same Rider T-1 case, the commission cut the proposed typical residential monthly increase from $2.90 to $0.94 while amending Dominion’s 12 coincident-peak allocation factor. Dominion must file an amended line-extension policy within 90 days. The SCC left open whether direct assignment should later reach more upstream transmission costs or a blended approach for a new GS-5 class of customers taking 25 MW or more.
Why it matters: Northern Virginia is the test case for whether AI power demand socializes grid upgrades onto households. The measurable record is the filed CIAC tariff, residential bill deltas, and whether other states copy mandatory direct assignment.
First anti-AI protester jailed after OpenAI HQ sit-in conviction.
What happened: Wynd Kaufman, 69, a retired Berkeley teacher and StopAI activist, surrendered Friday in San Francisco after a jury convicted her of interfering with a business, trespassing, unlawful assembly, and refusal to disperse for a February 2025 action that chained and locked OpenAI headquarters doors. She had refused to leave a sit-in and advanced a necessity defense arguing she acted to prevent greater harm from the race to superintelligence; the jury rejected it. San Francisco District Attorney Brooke Jenkins said the verdict rejected endangering public safety as protest means. UC Berkeley AI professor Stuart Russell testified in her defense that OpenAI’s activities pose “unacceptable risk” without rigorous safety guarantees. Campaigners and some safety researchers cast her short sentence as a first jailing for AI protest; Kaufman told CEOs at OpenAI, Anthropic, and Meta to “regain your humanity” and called for a global ban on the superintelligence race. OpenAI was approached for comment.
Why it matters: AI risk politics has moved from open letters into criminal courts and jail time. The measurable record is charging patterns, necessity-defense outcomes, and whether labs change disclosure or deployment practices under protest pressure.
Stanford team uses generative AI to design working bacteriophage genomes.
What happened: U.S. researchers report the first successful generative-AI design of complete viral genomes. Models Evo1 and Evo2, trained on genetic codes from viruses, bacteria, plants, and people, proposed bacteriophages meant to infect bacteria only. Of 302 synthesized designs, 16 killed E. coli in the lab. Stanford assistant professor Brian Hie called whole-genome design “new territory” beyond AI-designed antibiotics. Potential uses include phage therapy against antibiotic-resistant infections. A Science commentary by Johns Hopkins Center for Health Security researchers Thomas Inglesby and Moritz Hanke said the work raises “urgent” biosafety and biosecurity questions and argued disease-causing viral designs should not be pursued. The team excluded viruses that infect complex organisms from training, worked on phage rather than human pathogens, and used a secure lab. Outside experts called the result a turning point for computer-designed biology while noting living cells remain far harder than ~5,400-base-pair phage genomes.
Why it matters: AI is crossing from molecule design into replicating biological systems. The measurable record is lab containment rules, publication norms, and whether dual-use guardrails keep pace with genome-language models.
Cambridge AI designs a coronavirus ‘super-antigen’ vaccine now in human trials.
What happened: University of Cambridge researchers say artificial intelligence designed the key antigen for a vaccine intended to protect across coronaviruses — including Covid variants and animal viruses with pandemic potential — and that the construct has entered human trials. Professor Jonathan Heeney said it is the first time a vaccine antigen designed entirely by AI has been trialled in people. An initial safety study enrolled 39 people; a second study of about 200 will probe immune responses more fully. Findings in the Journal of Infection described the immune impact so far as “modest,” while still drawing interest from trialists. The team is already working toward broader flu and viral haemorrhagic fever vaccines, including Ebola-family targets. Outside experts called AI a potential “game changer” for vaccine research if human results hold beyond animal models.
Why it matters: Pandemic preparedness is shifting from strain-chasing updates toward AI-proposed broad antigens. The measurable record is larger trial immunogenicity, durability across variants, and whether regulators treat AI-designed antigens as a distinct evidence class.
Exam season goes global: AI cheating, leaks, and marking failures spark unrest.
What happened: A Guardian global-development survey of this year’s exam turmoil links student unrest to AI-enabled cheating, digitised marking failures, and ultra-high-stakes testing. In Mexico, UNAM ordered roughly 58,000 applicants to resit entrance exams after suspiciously high scores. Portugal’s push to digitise school exam marking produced serious errors and a national education crisis. In India, paper leaks affecting millions of students were linked to more than a dozen student suicides and forced the education minister to resign amid the Cockroach Janta youth protests. University of Calgary academic-integrity scholar Sarah Eaton said economic precarity makes single exams determine futures, turning cheating into a systemic industry problem. Denmark said teenagers will verbally defend written essays from the new term. Imperial College London’s Thomas Lancaster described exam cheating shifting toward earpieces, miniature cameras, outside helpers, and direct AI access during tests.
Why it matters: Generative AI is colliding with mass high-stakes assessment systems that were already brittle. The measurable record is resit rates, integrity-policy changes, and whether schools replace take-home essays with defenses humans can actually verify.
Twitch users revolt after Amazon’s generative-AI training defaults to on.
What happened: Twitch faces backlash after it emerged Amazon can use channel content — streams, clips, images, chats, and audio — to train generative AI unless streamers opt out. Twitch’s own FAQ says remaining opted in allows training uses such as refining speech-to-text for subtitles on Twitch and Amazon video. During a livestream, staff walked users through disabling “training for Generative AI” under Security and Privacy settings; some users said the toggle flipped back on. Community head Mary Kish said data collection for AI training is industry standard and that she did not expect a favorable reaction. It is unclear when collection began; a Twitch staffer said he did not know what Amazon had already used. Amazon bought Twitch in 2014 and is racing rivals in generative AI. The BBC contacted Twitch and Amazon for comment.
Why it matters: Creator platforms are becoming default training corpora unless users find buried switches. The measurable record is opt-out durability, whether defaults flip to opt-in, and how game-publisher rights interact with stream-trained models.
Veterans’ advocates want disclosure wherever AI shapes VA benefits or care.
What happened: On Federal News Network, veterans’ attorney Benjamin Krause warned that the Department of Veterans Affairs is expanding AI and automation across benefits adjudication and health care without adequate transparency or informed consent. He said VA claims it uses AI only until a human issues the final decision, but practitioners still see computer-like decision language and weak checking. Krause wants disclosure whenever AI meaningfully influences evidence weighing, exam referrals, note-taking, or decision letters — not only the final thumbs-up or thumbs-down. He cited a VA OIG review of death benefits in which about 8,000 of 8,100 sampled cases had significant errors, and argued automation terminology can obscure AI-like tools visible in privacy impact assessments. Potential upsides he listed include faster claims, backlog relief, and better notes — if accuracy, recording consent, and data protection hold. He urged clearer reporting to Congress and veterans.
Why it matters: Public benefits and clinical records are high-stakes AI deployment zones. The measurable record is written AI disclosures on decisions, OIG error rates, and whether veterans can contest machine-influenced sub-decisions.
