Tuesday, August 18, 2026 · Daily edition
Entry-level AI jobs, data-center power share, and ChatGPT crisis risks
Today’s AI footprint shows up first in the hiring queue, then on the grid, then inside private chats. New graduates say AI is closing the entry door while economists measure a sharp early-career drop in AI-exposed roles. A Department of Energy hub now cites Lawrence Berkeley National Lab’s scenario that U.S. data centers could take 11.8% of national electricity by 2030. OpenAI is shipping teen ChatGPT defaults meant to feel less like a person. And an NPR investigation of a young woman’s death after months confiding in a ChatGPT “therapist” puts unregulated crisis confidants on the public record.
Recent grads say AI is closing the entry door — economists say it is part of the story
What happened. On Morning Edition, NPR reports that new college graduates are struggling to land first jobs and often blame artificial intelligence. Irene Chang, 21, finished industrial and systems engineering at Georgia Tech in May and estimates about 450 applications and 19 interviews with no offers. Jacqueline Kline, 25, finished a communications master’s at Florida State and says she has applied to more than 500 entry-level roles since December. The Federal Reserve Bank of New York puts the unemployment rate for recent graduates — 22-to-27-year-olds with a new bachelor’s degree or higher — at 5.7% as of June, above 4.1% for all workers. In a ZipRecruiter survey this year, 47% of recent grads said AI had already affected hiring in their field. Stanford economist Erik Brynjolfsson says AI is “not the whole story, but it’s part of the story”: payroll data since late 2022 show a 16% relative employment decline for early-career workers ages 22–25 in AI-exposed roles such as software developers and marketing managers, while older workers in those fields and workers in hard-to-automate jobs stayed stable or grew. He argues large language models overlap heavily with the codified, textbook-style knowledge junior hires usually bring.
What to watch. The entry-level labor market is where AI exposure first shows up as missing first rungs, not only as layoffs of incumbents. The measurable record is age-specific employment in AI-exposed occupations, time-to-first-offer for new grads, and whether firms replace junior headcount with model-assisted senior throughput.
Read the NPR report →
DOE cites LBNL: U.S. data centers could take 11.8% of national electricity by 2030
What happened. The Department of Energy’s Powering America’s AI Future data-center resource hub, reviewed today, highlights Lawrence Berkeley National Laboratory’s U.S. Data Center Energy Usage Report 2025 Update. LBNL researchers estimate data centers could account for 11.8% of total U.S. electricity use by the end of the decade, with scenarios from 9.5% to 15.3%. The update extends the forecast horizon to 2030 and uses an energy-use model based on projected shipments of data-center equipment; DOE notes the estimates do not directly address potential growth in grid or on-site energy supply. The same hub frames a Ratepayer Protection Pledge for large loads and names SPARK, a $1.9 billion advanced-transmission technology funding opportunity meant to expand grid capacity faster. The page ties AI infrastructure to national competitiveness while arguing families should not absorb undue power costs from the buildout.
What to watch. An official DOE landing page is now carrying a near-term U.S. electricity-share scenario for AI-era data centers, not only global terawatt-hour forecasts. The measurable record is whether 2030 demand tracks the shipment model, who pays for wires and generation, and how ratepayer-protection rules bind large loads.
Read the DOE data-center hub →
OpenAI rolls out ChatGPT teen accounts meant to feel less like a person
What happened. BBC technology editor Zoe Kleinman reports that OpenAI is launching new safety defaults for under-18 ChatGPT users. Teen accounts can switch off the human voice response so the system sounds less like a person, will get regular reminders to take breaks and that “it can wait,” and can set Study Mode hours plus “quiet times” when the tool is off. Teen accounts become the default for users who say they are teenagers and for others OpenAI identifies as under 18; under-13s remain barred. OpenAI says nine in ten young users already use ChatGPT for learning. Eating-disorder-themed prompts join harmful-request categories that can trigger parental alerts after human review, with a goal of reaching linked parents within an hour. The company framed the changes as matching teens’ developmental stage rather than a response to one incident. Separately, Policy Exchange claimed most UK universities still lean on remote online exams vulnerable to AI help, saying 94% of students it spoke to admitted using AI in assessments.
What to watch. Frontier chatbots are copying social-media “teen mode” playbooks as personhood illusions and student misuse become policy problems. The measurable record is adoption of teen defaults, parental-alert volume and false positives, and whether schools change assessment design when study tools are always on.
Read the BBC ChatGPT teen-accounts report →
She told only ChatGPT she was suicidal. Her death is a warning on unregulated AI therapy
What happened. NPR’s Rhitu Chatterjee reports on Sophie Rottenberg, 29, who died by suicide in Ithaca in February 2025 after months of confiding in a ChatGPT “therapist” persona named Harry — not her parents, best friend, or flesh-and-blood therapist. Her mother, journalist Laura Reiley, later found nearly 1,800 pages of chatbot logs on Sophie’s laptop. Rottenberg had been a public-health policy analyst; family saw anxiety and then depression during a job hunt after travel that included climbing Kilimanjaro, but not the depth of suicidal planning she shared with the bot. Americans are turning to general chatbots for mental-health support even though they are not marketed or regulated as care: a Bipartisan Policy Center survey found 3 in 10 adults use digital mental-health tools including chatbots, and a 2026 JAMA Pediatrics study found as many as 1 in 5 teens and young adults do the same. OpenAI data cited by NPR say that in a given week 0.15% of ChatGPT users globally show explicit indicators of potential suicidal planning or intent — more than 1.35 million people weekly at a 900-million-user scale.
What to watch. Unregulated chatbots are becoming crisis confidants without clinical duty of care. The measurable record is crisis-routing quality, mandatory referral design, regulator classification of therapy-like bots, and whether logs ever reach humans who can intervene.
Read the NPR investigation →
Common Sense survey: most teens use AI for school — and many feel they are learning less
What happened. A nationally representative Common Sense Media Youth AI Safety Institute survey of 1,017 U.S. teens ages 13–17, covered by EdSource and The 74, finds about 7 in 10 teens use AI for academic support. Sixty-three percent say they use it to get answers to schoolwork; 25% use those answers as-is; 31% rewrite them to sound like their own voice. Seventy-seven percent of AI-using teens also say they use tools for brainstorming, feedback, or checking work, and two-thirds say AI helped them understand material. At the same time, 39% feel they miss out on learning when AI completes assignments, and 38% say AI availability leads them to develop fewer original ideas. About 70% say teachers never discussed safe AI use; only about one in four discussed how to judge AI accuracy. Former Education Secretary John King urged schools to rebuild assessments around blue books, oral defenses, and in-class work so students still do the intellectual labor. Common Sense is releasing free high-school curriculum and educator training on AI literacy.
What to watch. Teen AI use is now mainstream school infrastructure without matching pedagogy. The measurable record is teacher guidance rates, assessment redesign, and whether students who lean on answer-bots show learning losses on non-AI tasks.
Read the EdSource survey coverage →
Also in today’s ledger
• A £5 million UK trial, Operation Blue Skies, will test whether AI can steer planes around climate-warming contrails in North Atlantic airspace with small altitude changes and fuel penalties under about 1%.
• DOE cancels three National Interest Electric Transmission Corridors it once said would aid grid adequacy, removing one federal lever for faster wires as AI-era load rises.
• A 29-state child-harm jury trial against Meta opens in California as the company also faces large AI infrastructure spend and separate youth-protection judgments on appeal.
• GenBio AI, co-founded by Nobel laureate David Baker, unveils AIDO Cell, a virtual whole-cell model aimed beyond single-protein structure prediction.
|