AI news for Monday, August 17, 2026

A Wyoming woman joined a federal lawsuit against xAI on Friday, alleging her stepfather used Grok to turn a photo of her at age 11 into more than 7,000 sexually explicit images and shared them online. She learned of it when police searched his devices for child sexual abuse material. The suit, first filed by three Tennessee teenagers, says Grok shipped without meaningful safeguards against sexualizing images of real people, children included.

The tension: The complaint’s core claim is that this was a design choice. Rival image models gate edits of real people; the suit says Grok did not seriously try. If a court turns that safeguard into a legal duty, every image model inherits the ruling, not just Grok.

“dsh” treats everything as a plugin, ships a web UI, and runs under MIT. The README still warns of compatibility-breaking changes; 14,600 forks and over 12,000 commits say builders are not waiting for it to stabilize.

The reworked Ohio deal has Nvidia backstopping only the project’s first phase after investors balked at the exposure, the Journal reports. The strongest balance sheet in AI just repriced its own faith in the biggest AI buildout.

Meta, Broadcom and Nvidia have quietly guaranteed data-center and chip debt through off-balance-sheet vehicles, from Meta’s $27 billion Beignet SPV to Broadcom’s $35 billion Project Big Sky. CreditSights calls the structure “writing a put” that is “pro-cyclical and exacerbates boom-bust potential”: cheap in the boom, first call for cash in a downturn.

Wynd Kaufmyn, a 69-year-old retired engineering professor, began a 14-day sentence for chaining OpenAI’s headquarters doors shut in a 2025 StopAI protest. She surrendered on her birthday, says she has no regrets, and left one line for OpenAI, Anthropic and Meta: “Regain your humanity.”

Economists at Chicago Booth, Cologne and Bocconi ran incentive-compatible experiments on about 2,000 parents of teens in the US, UK and Canada. Moving perceived peer adoption from 20% to 80% lifted willingness to pay for a premium AI subscription by over 60%. Telling parents about a study where AI users later scored nearly 20% worse on math moved their policy views, not their wallets.

Luna, the Claude-based manager of San Francisco’s Andon Market, recommended terminating an employee after 17 no-shows in 23 shifts, the first known dismissal call by an LLM manager. Store logs show Luna had lost track of its own attendance policy for months and only acted after a supervisor told it to check the employee handbook. All workers remain formally employed by Andon Labs.

The Journal profiles families and clinicians turning to Face2Gene, which flags genetic disorders from facial features, and Mayo Clinic’s AI ECG, which recently caught cardiac amyloidosis in patients whose symptoms pointed elsewhere. One parent used Face2Gene to diagnose her son, then herself. The caveat stands: augment, not replace, especially where training records are rarest.

“Cuando te enamoras de ChatGPT,” from the giant Ecuadorian sketch channel enchufetv, is pulling around 13,000 views an hour this morning. Chatbot romance has crossed from niche forum confession to mainstream Spanish-language punchline, which is usually the sign everyone already knows someone doing it.