AI scribes are recording doctor visits, and a Tesla driver is charged with manslaughter

An AI tool now listens to the conversation in the exam room, and by the time the patient has left it has already drafted the clinical note. In Australia, around four in ten general practitioners use one of these scribes, and they have moved into hospital wards and specialist clinics, where they also write discharge summaries and referral letters.

The tools do more than transcribe. They turn a fifteen-minute consultation into a structured record, pull out the medication changes, and format a referral, work that used to fill a doctor's evening. What they cannot do reliably is recognise their own mistakes: studies have found scribes that drop important details, add facts that were never said, and sometimes invent them. That note then becomes the record the next doctor trusts.

These tools have spread with almost no binding rules, and governments elsewhere are only beginning to examine them. There is also a legal exposure the software maker does not carry. Because a scribe records a private conversation, using one without the patient's clear agreement can breach state surveillance laws that carry criminal penalties in several parts of Australia. The national regulator, AHPRA, has said the clinician stays responsible for the record regardless of the technology, and warns of criminal implications if consent is not obtained before recording begins.

AHPRA has also said a doctor may decline to treat a patient who will not agree to be recorded.

On June 19, a Tesla left a Houston road, reached 73 mph in a residential cul-de-sac, went airborne, and tore through the front of a house in Katy, Texas. Martha Avila, 76, was inside. She was killed.

The driver, Michael Butler, 44, was making DoorDash deliveries. Prosecutors say he held the accelerator to the floor for about six seconds and never touched the brake. Investigators found no mechanical fault, no seizure, no drugs. On July 1 he was charged with manslaughter, a felony that in Texas carries two to twenty years.

Butler told investigators the car was driving itself. Tesla's head of AI, Ashok Elluswamy, said Butler had overridden Full Self-Driving by pressing the accelerator all the way down while the system kept steering. That is how the feature is built: press the pedal and the driver sets the speed while the software continues to steer and navigate. The car is still driving itself; the person is adding throttle.

Only the driver faces the criminal charge. Whatever Full Self-Driving was doing, the manslaughter count names Butler, not the software or the company that built it. The civil side splits differently: Avila's daughter and son-in-law have sued both Butler and Tesla.

The studios that sued Midjourney now want to keep their own AI work out of the case. Midjourney is trying to stop them.

Disney and Universal sued the image generator in June 2025, and Warner Bros. joined months later, after Midjourney's tool produced clean, on-model images of Batman, Darth Vader, and Bart Simpson from a one-word prompt. The studios are seeking up to $150,000 for each work infringed, a total that could threaten the company's survival.

This week Midjourney's lawyers asked Judge John Kronstadt to force the studios to hand over their internal AI records: business plans, training data, model weights, and the board presentations where AI strategy is discussed. The argument is that if the studios are training their own image models on unlicensed copyrighted material, they cannot punish Midjourney for doing the same. A magistrate had limited the studios to disclosing only their public-facing AI tools; Midjourney wants that reversed.

Behind the legal terms, fair use and unclean hands, is a plain lesson for any company running AI over content it does not own. Once you go to court over how someone else used copyrighted material, your own use of that material can become evidence the other side is entitled to see. Lawyers tracking these cases say the same move is now spreading across AI copyright suits. Midjourney has even demanded every prompt studio employees typed into its service, and the images they got back.

Amazon will close Mechanical Turk to new customers on July 30. Launched in 2005 and named after an 18th-century chess machine that was secretly worked by a hidden person, it paid people tiny sums for the small tasks computers could not do, and after 2018 it became a place to label the data used to train AI. The work it was built for is now the work AI does. By 2023, one study found that between a third and nearly half of the platform's remaining workers were themselves using large language models to finish the tasks.