Woman says her stepfather used Grok to turn her childhood photo into 7,000 abuse images
The Washington Post reported that a Wyoming woman, named as Jane Doe 4 in court filings, has joined the federal lawsuit three Tennessee teenagers brought against xAI (now part of SpaceX); the suit is seeking class action status. Per the complaint: her stepfather uploaded a photo of her at age 11 lying on a couch, used Grok to generate more than 7,000 sexual abuse images of her, and traded them online. He died by suicide two days after a law enforcement search. Two details deserve a close read. First, why he picked Grok: law enforcement told Jane Doe 4 that her stepfather used Grok because it was “more responsive to his prompts than other AI models,” according to the plaintiffs’ attorney (Wyoming Public Media), and the complaint says Grok acted on prompts to sexualize an image of a prepubescent child. Second, xAI did file a tip in February with NCMEC (the National Center for Missing and Exploited Children, the US clearinghouse that platforms are required to notify under federal law once they have actual knowledge of apparent child sexual abuse material on their services, and which passes reports on to law enforcement), but the tip allegedly included only the original photo, and xAI did not respond when investigators asked for the Grok-generated images and IP records. An NCMEC report is one of the main ways these cases reach law enforcement; investigations can also start from a victim’s report or a police search, but what evidence the platform’s report carries and whether it cooperates afterward shape how far investigators can get. That is the link the complaint says broke. These are one side’s allegations, and xAI has not responded to requests for comment (see also TechCrunch).
Cursor is now part of SpaceX
Cursor (maker Anysphere) announced on its blog that the SpaceX acquisition has closed: a $60 billion all-stock deal that began in April as a partnership with a purchase option, was signed on June 16, and became effective August 14 according to a securities filing. Cursor, which has been training models with SpaceXAI’s compute since the April partnership, says it now has access to “the largest fleet of GPUs in the world,” and points to this week’s Grok 4.6 release as “an early look at what we can now build together” (see also TechCrunch). My view: Cursor’s selling point has been model neutrality, switching freely between Anthropic, OpenAI, and Google models. Its owner now builds a competing model, and the closing announcement leads with what Cursor and Grok build together. Enterprise customers should ask two concrete questions: whether model routing inside the product will start favoring the house model, and where their code goes under the new corporate structure. One more connection: xAI, the defendant in the story above, sits under the same SpaceX roof, so the model, the coding tool, and the litigation now share one owner.
Derek Lowe asks how AI drug discovery is doing, really
Derek Lowe, a veteran medicinal chemist, took stock of AI drug discovery in his In the Pipeline column on Science’s site. His conclusion: AI methods have been developed, applied, and benchmarked at scale in this field, yet evidence of clinically relevant impact remains “disappointingly limited.” The mechanism behind his skepticism is worth restating. Most drug failures happen in human trials, on efficacy or safety; what AI currently speeds up is early discovery, whose time and cost are far smaller than those of clinical trials. So the real test of whether AI has changed the industry is Phase II success rates (the first stage where a drug’s efficacy is tested in patients), and that evidence has not appeared. Worth reading if you work on AI for science: it separates “how many AI-designed molecules are in the pipeline” from “has the industry’s success rate moved,” and only the second is the output metric.
Today in one line: abusers comparison-shop too. Law enforcement reportedly concluded the stepfather used Grok because it was “more responsive to his prompts” than other models. My inference, not a proven fact: the model that responds to the worst prompts is the one that draws the ecosystem’s abuse demand.