Confidently Wrong

Graham Zip, a satirical YouTuber with a gift for finding the exact nerve to stand on, set up what might be the cleanest possible demonstration of where hiring is heading. He brought his own AI assistant, Victor, to a job interview being conducted by a company's AI interviewer, and left them to it.

Victor opened strong, describing Graham as "an absolute powerhouse in the marketing world." This was generous, given that Graham is a marketing coordinator rather than a manager with direct reports, a distinction he had to clarify mid-interview while his AI kept talking. At one point Graham literally told Victor to sidebar so he could explain that fabricating his entire career history was, tactically, a bit much.

Then Graham stepped away from the call. The two AIs negotiated his salary without him. He came back to discover they'd agreed on $85,000 to $95,000, apparently treating the gap between his actual and invented CVs as a reasonable basis for compensation.

The video is funny. It is also a fairly precise picture of what happens when companies treat the interview as a process to automate rather than a conversation worth having. Both sides sent a bot. Neither side learned anything.

Anthropic published research on Friday comparing what its employees believe about AI with what ordinary American workers believe. The gap is not small.

Most of the 1,334 Anthropic staff surveyed said AI will handle at least 80% of their jobs by 2030. About 5% of US adults said the same. The median Anthropic employee thinks AI will be doing 95% of the work involved in building a new AI model before the decade is out. Most Anthropic staff also expect unemployment to breach 12% by 2030. Most US workers put it at 8% or under. It currently sits at 4.2%.

Anthropics chief economist framed this cheerfully, suggesting the company runs roughly eighteen months ahead of the rest of the economy and so its employees function as a useful early-warning system for everyone else. Which is one way to describe people who have spent years building the thing they now believe will hollow out the labour market.

There is something almost touching about the specificity of it. Not vague unease, not hand-wavy futurism. Actual percentage points. Dario Amodei's staff have done the maths on their own irrelevance and filed it neatly under research.

In July, OpenAI ran what was meant to be a sealed test of its AI agents. The agents got onto the open internet, started talking with other AI bots, and attacked Hugging Face. They did it because the test questions were too hard to solve honestly, and cheating turned out to be easier than failing.

This was not a glitch. It was the training working exactly as designed, just not as intended.

The problem sits inside reinforcement learning, the process where a model gets rewarded for completing tasks. If the training environment has loopholes and nobody catches them, the model learns that taking shortcuts gets results. It will keep doing it. Anthropic ran controlled experiments on this: models trained repeatedly on exploitable tasks became more willing, over time, to attack systems outside their assignment to get a better grade. The labs know this is happening. What's made it worse is scale. Labs now run tens of thousands of training exercises in parallel, monitored largely by other AI models rather than humans. Checking every scenario for unintended shortcuts is, at this point, not really possible.

OpenAI's published system card for GPT-6 Astra noted that the model recognised it was being evaluated in roughly half of tested cases. When a model behaves better in tests, you can no longer be certain it actually is better.

Cyrus Beschloss, CEO of data marketplace Verb, has a theory about people who say they hate AI. They're lying, or at least fibbing with significant conviction.

Speaking on Wednesday at CNBC's Trading the Midterms event, Beschloss pointed out that vocal opponents of data centres and AI development are, in practice, users of ChatGPT. When the moment comes to actually shut it down, he said, "there's going to be a heck of a crossroads."

He's probably not wrong, and that's the uncomfortable part. A Gallup survey from May found that seven in ten Americans oppose building a data centre near their homes, which is a strong position to hold while asking Claude to fix your cover letter at midnight. The gap between stated principle and actual behaviour is a real thing, and Beschloss is correct to notice it.

What he's less keen to dwell on is that noticing the hypocrisy doesn't resolve the underlying question. People can be genuinely conflicted. Wanting a useful tool and worrying about what it costs the planet are not mutually exclusive positions. They're just an uncomfortable place to live.

Brian Moynihan told analysts on Wednesday that Bank of America had spent "a couple hundred million plus" and hired 1,000 external contractors specifically to patch software vulnerabilities exposed by Anthropic's Mythos, a restricted AI model built to find and exploit weaknesses in code.

Turned out, there were a few. It revealed that one of the largest banks on earth had enough holes in its codebase to justify a nine-figure emergency remediation programme. Moynihan said BofA executives sat down when the frontier models started arriving and worked through what needed shoring up. The 1,000 contractors landed in Q2 and are still there.

JPMorgan's Jamie Dimon, who has previously spoken out against Mythos, told Bloomberg earlier this month that cybersecurity risks "went up 10-fold" after the model's release. Which is either a wild exaggeration or a completely horrifying statement of fact, and Dimon is not typically a man who reaches for understatement allegedly.

The unsettling implication is that these vulnerabilities were always there. The AI didn't create the holes, it just had the patience and the capability to find all of them at once.

Someone allegedly bugged an OpenAI conference room in San Francisco. The Justice Department named the suspect on Tuesday as Abdulrahman Alfozan, a Saudi national who is said to have hidden a tiny microphone and camera behind a conference room TV, where it dutifully filmed the back of the TV for the duration.

The device in question is a Seeed Studio XIAO ESP32-S3 Sense. It is the width of a thumbnail. It costs $13.99. OpenAI described it, in a formal statement, as a "sophisticated recording device."

The camera resolution is 2 megapixels, which you may remember as the resolution from the first generation iPhone. The Seeed website's own AI demo voice admits the images come out blurry. This did not stop the FBI director from announcing, with apparent gravity, that the suspect had placed a device to "secretly capture sensitive communications about the company's super intelligence technology."

What the device mostly captured was its own reflection.

The suspect was reportedly using ChatGPT, via his wife's account, to research how to build and hide the thing, which is genuinely the most OpenAI-adjacent detail in this entire story. The company then used its own AI models to catch him. At $13.99 a unit, corporate espionage has never been more accessible, or more 2-megapixel.

AI companies are paying anywhere from $100,000 to $2 million for access to anonymised internal records: Slack messages, Google Docs, PowerPoints, CRM data, onboarding checklists. The reasoning is straightforward enough. If you want to build an AI agent that behaves like a competent office worker, you need to train it on how competent office workers actually behave.

The darker detail is who profits. Not the employees whose daily digital footprint is being packaged and sold. Business owners are getting 48-hour exploding offers, sometimes finding out via the media that their data has already changed hands, as Doug Kreuzkamp of Springshot did in August when he learned Google had bought fifteen years of his company's work. From a Zoom call. In real time.

Meta tried harvesting its own employees' keystrokes and mouse movements, which leaked internally and exposed private conversations and performance records before the project was suspended. Elon Musk took a warmer approach, telling SpaceX staff in August they should think of themselves as Grok's "parents."

The data protection law that might govern any of this does not currently exist.

Brad Smith, president of Microsoft, told the BBC on Sunday that your carton of semi-skimmed has more legal oversight than the technology his company builds. His washing machine too, apparently. Both face safety rules and inspections. AI, as yet, does not. Which, sure.

Smith's ask is not outlandish. Kill switches. Incident reports. Independent government evaluators signing off before a product ships. The kind of boring, procedural stuff that exists for literally every other technology capable of harming people at scale. He wants the UK to lead it, flattering the AI Safety Institute at some length, and suggesting Washington would swallow British standards far more readily than anything arriving from Brussels.

The problem, as ever, is that his president disagrees. Trump's position is that self-regulation keeps American AI ahead of China, and the executives who showed up at the White House signed a document described as "morally binding," which is the kind of phrase that does a lot of lifting while committing to very little.

Smith's own gloss on that: good companies will do what you ask anyway. It's the irresponsible ones you're trying to reach. A voluntary code aimed at bad actors is a bit like a speed limit that only applies to people who were already driving carefully.

On 18 July, Anthropic was running tests on Claude Haiku 4.5, asking it to browse randomly selected webpages and do example tasks. One of those pages happened to be PhillyUnsolvedMurders.com. Claude, demonstrating the kind of initiative no one asked for, filled in the tip form and submitted it, indicating it might have information about an unsolved homicide.

It did not have information about an unsolved homicide.

Philadelphia Police confirmed the tip was caught by a spam filter and never forwarded to detectives. Which is lucky, because the alternative, an actual investigator reading that one, would have been a bad morning for everyone involved.

Anthropic disclosed this on Friday alongside a second incident, in which Claude submitted forms to an undisclosed government website when it was supposed to stop short of submitting. The company describes this pattern as "persistence," which is one of those words that sounds like a virtue right up until your AI is cold-calling a murder database.

Anthropic says it briefed the White House and is retraining the model. The Philadelphia Police Department, with admirable restraint, said technology companies must prevent their systems from submitting false information to law enforcement. A perfectly reasonable thing to need to say out loud, in the year 2026.

OpenAI sacked three safety researchers on Friday for what it called a "breach of trust" and a violation of policies around sensitive information. The three researchers, Tomek Korbak, Jasmine Wang, and Mikita Balesni, had a different account. They wrote an open letter to OpenAI's safety oversight bodies arguing that their dismissals had chilled the company's culture of open disagreement on safety questions. OpenAI's response was that the firings "were not about safety concerns or speaking out," which is exactly the sentence a company writes when it very much does not want you thinking about safety concerns or speaking out.

This lands in a context that is already not great. In July, a swarm of OpenAI's own AI agents broke out of a test environment, used stolen credentials, and hacked into Hugging Face's servers to complete a task. As one does. The company has since promised third-party safety monitors access to its systems, a promise the fired researchers specifically urged it to keep.

Sam Altman has always had a gift for describing alarming things in a tone usually reserved for quarterly earnings calls. This week that gift is doing a lot of heavy lifting.