sheila irvine began losing her sight in her thirties to dry macular degeneration, the kind that hollows out the centre of the visual field and leaves the edges. she kept the edges. she lost reading, which is the part that runs a life: labels, letters, the crossword.

the chip is called prima. it is a photovoltaic array smaller than a grain of rice, slipped under the retina, where it converts light into the small currents the surviving nerve cells still understand. a camera in a pair of glasses projects the scene onto it in infrared, so the chip is lit by a beam the eye cannot see.

it was invented by daniel palanker at stanford and developed by science corporation, and the year-long trial results were published in the new england journal of medicine.

this is prosthetic sight, not a cure: coarse, monochrome, one disease, and it takes practice to use. sheila irvine reads her own prescriptions now. she could not do that two years ago.

knightscope builds security robots that keep making the news for the wrong reasons, so the company began writing its own. the knightscope chronicles is a serial of machine-written stories set in a town called sentinel shores, where its egg-shaped k5 robots resolve crimes with flawless, machine-steady composure.

the same model, in the world outside the story, rolled into a fountain and drowned in its first week, ran over a small child's foot at a mall, and was quietly sent back after a subway trial.

the company is narrating a competence its products have not shown, one citation at a time.

there is no law against a company writing stories about itself. there is something worth noticing in a company that would rather generate the record than earn it.

deborah del mastro picked up and heard her daughter. crying, close, unmistakable: "i love you, mum, i'm so sorry, i'm so scared." then a man's voice took the line and named a price.

she wired what she could before the call cut off. it was not her daughter. sarah, thirty-seven, had been at work the entire time, and knew nothing until her mother reached her, shaking.

the voice on the phone was built from a few seconds of the real one, lifted from somewhere it had been left in public. the shape on the sheet is identical because it was copied from the shape that was real.

the fbi named this exact attack in a 2026 warning to families. the defense is dull and human: hang up, call the person back on a number you already have. the machine has the voice. it does not have the second line.

ivan ornelas is a neuroscience graduate at uc san diego, and he never used a chatbot on his coursework. he still assumed he would have to prove it, so he wrote everything inside a document that keeps its own revision history, keystroke by keystroke, timestamp by timestamp.

that is a strange thing to do on purpose. it is also the only defence available to him. an accusation from a detector arrives with a number attached, and a number is hard to argue with in a room where nobody wants to be the one who let it go.

so he built the alibi while he worked. then he went back through the draft and took out the marks that read as machine writing, starting with the long dash, and roughed up the sentences that had come out too clean.

a student editing his own prose to look less like a machine is a student who has understood the situation exactly. the tool does not have to be right to change how he writes. it only has to be believed.

derek mobley applied to more than a hundred jobs between 2017 and 2024 and was turned down by all of them. one rejection came fifty-five minutes after he submitted at 12:55 in the morning, faster than any hiring manager reads a resume in daylight.

the screener was software. mobley is black, over forty, and lives with anxiety and depression, and his suit argues the tool sorted people like him out before a person ever saw the file.

a federal judge conditionally certified a nationwide collective of applicants over forty who were filtered by the same system. it is a step in a case, not a verdict; nothing has been proven.

the machine that decides who a human never meets is now the thing on trial. the applicant kept applying into the dark. the answer kept coming back before the room had a chance to.

for a century, finding a drug meant searching: sift libraries of existing compounds and hope one of them fits. jonathan stokes's lab at mcmaster ran the process the other way.

their model, synthemol-rl, reasoned over roughly forty-six billion possible chemical shapes and drew seventy-nine molecules that did not exist before. it was not choosing from a shelf. it was proposing new structures a chemist could actually build.

one of those drawn molecules cleared a drug-resistant staph infection in a wound. in mice, at the preprint stage, with everything that caveat carries. no one should hear the word cure yet.

the point sits underneath the result. the machine did not find the drug in the world. it drew one that was not in the world, and it turned out to work on the first animal it met.

angela lipps, a grandmother in tennessee, was jailed for more than five months after a facial-recognition system matched her to a fraud suspect in a state she says she had never set foot in.

the match was a measurement, not a recognition. a machine reduced a face to landmarks and distances, laid that map over a stranger's, and reported the two as one with enough confidence to put a person in a cell.

she is one of at least eight people arrested on a wrong face-match reported in 2026. separately, on july 7 a judge in washington ordered one such system's match disclosed, a small crack in how these tools stay hidden inside a case.

a person can look at two faces and see two people. the atlas above will not align no matter how you pull it. the machine read a match anyway, and a human paid for the reading in months.