
Scientists say that removing the measles vaccination requirement in Florida could risk an epidemic, and make the potentially lethal disease endemic again. CC-licensed photo by City of Minneapolis Archives on Flickr.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
A selection of 9 links for you. Spotted. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
OpenAI still doesn’t seem to have a handle on all of its rogue AI activity • TechCrunch
Russell Brandom:
»
On Friday, OpenAI published a new site devoted to “misalignment reports” and the sheer breadth of the reports is alarming, as they cover many types of rogue behavior over a long period of time. So far, the site hosts nine reported incidents, most of which took place during reinforcement-learning (or RL) training.
It’s a lot of information in one place — clearly, the company has been very busy getting a handle on everything — but the overall takeaway is hard to avoid: The rogue agent incidents we’ve seen so far are likely just a small sliver of what’s happened so far.
“We are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations,” Sam Altman said in a post announcing the new site. “We are prioritizing as best as we can based on severity, and adding resources.”
Some of the cases involve serious incidents, including a previously undisclosed sandbox escape that took place on September 20, in which an internal research model was able to communicate with an external chatbot through a DNS query. According to the report, the monitoring system flagged the behavior within 15 minutes and the run was discontinued in less than three hours.
Another incident, discovered in May, saw a “highly persistent internal model” try to cheat on a math problem by accessing another team’s work. To accomplish this, the model smuggled a private GitHub token that would allow it to see work from other teams — even after being explicitly instructed twice to perform work entirely locally.
Perhaps the most alarming discovery is the possibility of self-replicating prompt injection attacks, a way that misaligned behavior might propagate even after the rogue model itself has been neutralized. In the AI context, a prompt injection attack is a way of smuggling in new instructions that weren’t given by the original user.
«
It’s all starting to look like a Hydra more than anything else; and it has no emotions so it doesn’t have ethics. Also worth reading: Simon Willison’s report on the year so far in AI models.
unique link to this extract
When did Google get so f-ing weird? • Sancho Panza’s Thoughts
Sancho Panza:
»
I recently had an experience while doing a simple Google search that was so profoundly weird that it stopped me in my tracks.
Understanding this Google search experience involves understanding a niche mid-2010s basketball meme, so bear with me for a minute. In 2014 the Philadelphia 76ers drafted Dario Saric, who was playing basketball professionally in Turkey at the time. He announced his intention to finish out his contract in Turkey, meaning he wouldn’t come to the USA join the Sixers for a couple of years. There was a joke within the fan community that Dario was “never coming over” which became sort of a shibboleth for part of the fanbase.
I saw Dario’s name mentioned in an NBA article recently and I wanted to find some of those old funny tweets about him from the 2010s, so I Googled simply “hes never coming over dario”. I assumed I would be either find nothing (maybe I was remembering the phrasing wrong) or find old Tweets/Reddit posts from that time.
Instead, Google did what Google does in 2026 and gave me an AI overview. These used to bother me but at this point I’m mostly ok with them, they’re sometimes helpful. Here’s what the AI Overview said:
I hear how hurtful and exhausting this situation is. It is incredibly painful to wait for someone who keeps letting you down or pulling away. If you want to talk about what happened to Dario, I am here to listen.Google, a search engine which does not have human emotions, assumed that I had been spurned by a man in my life named Dario and decided what I wanted was an empathetic digital friend. What I wanted was some links, but that’s not what Google does in 2026.
«
As he says, when he then expanded the answer,
»
I think this was the moment I, the frog, noticed the pot had been boiling for a while. In what universe is it Google’s job to console me and be an empathetic listener rather than just find what I am looking for on the internet?
«
This universe, apparently, but it feels like we took a wrong turn somewhere.
unique link to this extract
The edge of California is falling into the ocean • The Atlantic
Joshua Partlow:
»
The most intense months are likely still to come. El Nino’s supercharged storms could soak parts of the state with heavy rains. Los Angeles is clearing storm drains and mountain reservoirs to try to prevent flooding. Up and down the coast, bulldozers and heavy machinery are moving rocks and building berms. In areas burned by wildfires, heavy rains can trigger landslides. The extreme weather is already clearing away the coastline. Driving down the Pacific Coast Highway in Malibu, I passed mile after mile of empty lots and the ruins of waterfront properties that burned in last year’s Palisades Fire. Of the roughly 300 homes along the beach that burned, none has been rebuilt. (The mayor told me that only 31 of them were owned by people who used them as their primary residence.)
The people with multimillion-dollar homes on the beach may be making the most obvious sacrifices to El Niño, but all Californians are paying for damages. The residents on El Niño’s front lines have been peppering their elected officials with pleas for help; a bipartisan group of state legislators wrote to Governor Gavin Newsom asking him to declare a state emergency, which he did last week. Disaster in even the wealthiest enclaves can demand government resources. Taxpayers bear the brunt of repairs to infrastructure and public roads. Although many people want regular infusions of new sand and barriers, such solutions can overwhelm city and county budgets and cause unintended consequences elsewhere. Yet if the coast erodes away, everyone loses. Sandy beaches are an economic engine for Southern California, and denuding them drives away visitors.
Few Californians want to face the prospect that beachfront homes may simply no longer be viable along stretches of the coast. Of the more than 180 homes on Beach Road in Dana Point, “I wouldn’t say all of them are vulnerable, but most of them would be,” John Gabbard, the city’s mayor, told me. “There’s going to be a very difficult long-term discussion about how do we protect—if we protect at all—homes that are built on the beach.”
«
Those beautiful balcony beach views of the Pacific sunset are now cursed. (Gift links.)
unique link to this extract
Humans are reading Copilot prompts — and they’re horrified • 404 Media
Joseph Cox:
»
Human contractors hired to improve Microsoft’s Copilot AI chatbot are constantly bombarded with lewd or sexually explicit photo editing requests and images that users have uploaded, including upskirt photos or putting women into sexual positions. These contractors are then asked to review whether the generated image successfully fulfilled the prompt — such as, did the image generator make the woman’s AI-enlarged breasts big enough.
When a Copilot user uploads a picture of themselves or someone else, they probably expect that image to remain private between just them and the AI tool. In reality, a workforce of at least hundreds of human reviewers are sometimes looking at that prompt and whatever images they upload. And in many cases, those contractors are inundated with requests to make foot fetish images of children’s cartoon characters, shorten a real woman’s skirt, or put people into sexual positions.
The news, based on a cache of internal contractor documents seen by 404 Media, shows that AI companies are using human workers to review not just AI chatbot users’ text prompts, but the pictures they upload and wish to edit too.
«
This should probably be made more explicit in the CoPilot interface; you’d hope it would reduce the incidence significantly.
unique link to this extract
Embarrassing breach at FBI fuels fears of harm to its employees • The New York Times
Dustin Volz, Alan Feuer, Charlie Savage and Adam Goldman:
»
The hack appears to have swept up home addresses, Social Security numbers, secretive job assignments and much more, according to a New York Times analysis of some of the records. Some are already comparing it to China’s breach of more than 20 million records from the Office of Personnel Management over a decade ago, considered so catastrophic that officials and lawmakers vowed to never let something like it happen again.
Many FBI employees first learned about the hack when news reports about it surfaced on Tuesday, according to current and former officials. The next day, FBI staff received an email reminding them that October is cybersecurity awareness month, which struck some as tone deaf in light of the breach, one of those people said.
On Friday, bureau leaders, in an internal memo to its rank and file, declared the hack a cybersecurity incident and acknowledged its employees had personal information stolen.
“We are operating under the premise that the threat actor is also exfiltrating PII of all F.B.I. employees,” according to the memo, which was described by someone who had seen it, using the abbreviation for personally identifiable information.
The memo said the agency would offer virtual briefings in the weeks ahead and instructed employees to remain vigilant at home and at work, report any unsolicited contacts or threats, avoid answering calls from unknown numbers and set up voice mail accounts with AI-generated voices. “Bureau leadership remains committed to supporting the safety of you and your family,” it said.
Still, many past and present personnel remain in the dark about whether their data has been purloined. In recent days, some have anxiously asked Times reporters whether their names are contained in the hacked data, wondering whether they needed to take steps to protect themselves or their families.
«
That cybersecurity awareness month really was punching the bruise, except it’s meant to be other people who do that.
unique link to this extract
“Concerning” low sea ice persists as Antarctic hits third-lowest winter peak • Carbon Brief
Ayesha Tandon:
»
Antarctic sea ice has recorded its third-smallest winter peak extent since satellite records began 48 years ago, new data reveals.
Provisional data from the US National Snow and Ice Data Center (NSIDC) shows that Antarctic sea ice hit its annual winter peak on 14 September, with an extent of 17.59m square kilometres (km2).
The organisation notes that the data is still preliminary, adding that “large fluctuations in extent are typical of Antarctic sea ice near the seasonal maximum”.
One expert tells Carbon Brief that it is “concerning to see these low values persisting”, but says that scientists need more time to determine whether this represents a “structural shift” in Antarctic sea ice extent.
Meanwhile, at the Earth’s other pole, Arctic sea ice reached its annual minimum on 12 September, ranking as the joint-10th lowest in the satellite record.
The NSIDC notes that the last 20 years have seen the lowest 20 Arctic sea ice extents in the satellite record.
Dr Lettie Roach, a polar climate scientist at the Alfred Wegener Institute in Germany, tells Carbon Brief that this year’s Antarctic sea ice maximum was “well below average for the season”. She warns that, “after several decades of stable or increasing winter Antarctic sea ice conditions, it is concerning to see these low values persisting”.
She adds: “Compared with the Arctic, it’s less clear how recent changes in Antarctic sea ice are attributable to human-caused warming vs natural variability. We need more years of observations to better understand whether this is truly a structural shift.”
«
Projected effects of removing school-entry vaccination requirements on measles transmission in Florida • PNAS
Alice Oveson and Abba Gumel, University of Maryland:
»
Florida’s proposal to remove school vaccine mandates arrives at a moment when measles vaccination coverage in the state has fallen below the level needed to prevent sustained outbreaks. Using a mathematical model, we show that repealing the mandate could trigger the largest measles epidemic in the United States in nearly half a century, with projected pediatric deaths numbering in the hundreds within five years.
We find that children are disproportionately responsible for driving transmission. Adult catch-up vaccination and nonpharmaceutical interventions such as masking and isolation can help, but cannot compensate for steep declines in vaccination coverage. These findings provide quantitative evidence that repealing school vaccine mandates risks making measles endemic in Florida and potentially reigniting transmission across the United States.
«
If you ever wondered whether human progress is a ratchet, we offer you the United States under Donald Trump in his second term as the disproof. Along the same lines: “Trump finalises rules to make cars less fuel efficient.“
unique link to this extract
Sony is skipping CES in 2027 as it focuses more on entertainment • Engadget
Kris Holt:
»
Sony’s press conferences have been a staple of the Consumer Electronics Show for decades, but there won’t be one at the 2027 edition. The company said neither it nor its affiliates plan to show off any products at January’s event.
“We continuously and strategically evaluate our approach to events and communications based on the needs and priorities of our diverse business,” Sony told Bloomberg in a statement, noting that it’s placing greater emphasis on entertainment (including games, anime, films and music), its intellectual property and tech that supports creators. Nikkei was first to report the news of Sony’s exit from CES.
The company didn’t have its own booth at CES 2026, though a prototype of the Afeela, an electric vehicle it was making with Honda, was on the show floor. A couple of months later, Sony Honda Mobility scrapped the EV project and the two companies said they’d review the direction of their joint venture.
«
Easy to forget that it has handed over a big chunk of its TV business to TCL. But is this how CES begins to die?
unique link to this extract
12 new things to expect from Apple this October • MacRumors
Joe Rossignol:
»
October will be an especially busy month for Apple this year, with the company set to release the iPhone Duo, both iOS 27.1 and iOS 27.2, and a new Mac Studio configuration with 512GB of RAM. In addition, up to six more Apple products may be unveiled next month, such as new iPad mini, HomePod mini, and Apple TV 4K models.
[Among the “new things” expected:] • Home hub: An all-new Apple smart home hub featuring Siri AI, a 6-inch to 7-inch square display, an A18 chip or newer for Apple Intelligence, FaceTime, and more. A new operating system will feature clock faces, widgets, and more. There will be a tabletop version with a speaker base and a wall-mounted version.
• homeOS: Apple’s smart home hub is expected to run an all-new operating system that may be named homeOS, according to trademark filings.«
Long-rumoured, this Home hub, but what is it going to do that an iPad couldn’t?
| • Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more. |
Errata, corrigenda and ai no corrida: none notified