In The Register

Again, these are not "hallucinations" These are the system working as designed and producing a long string of text that plausibly mimics the kind of text requested. The system is not designed to produce factually correct statements and will not, except by coincidence.

KPMG's October 2025 report on the wonders of agentic AI has been accused of demonstrating one of the tech's less desirable talents: making things up.

Research outfit GPTZero claims a forensic review of the Big Four firm's October 2025 report, "Total Experience: Redefining Excellence in the Age of Agentic AI," found that only five of its 45 citations correctly pointed to the cited source; the rest ranged from mangled and misleading to partially fabricated or too vague to verify.

[…]

GPTZero dubbed the phenomenon "vibe citing" – the citation equivalent of vibe coding – where generative AI appears to stitch together fragments of real sources, invent titles, or otherwise produce references that look convincing until someone actually clicks them.

GPTZero alleges that roughly half of the report's factual claims were false, unsupported, or attributed to the wrong source. Several case studies highlighting supposedly cutting-edge deployments of agentic AI appear to have been particularly creative.

In calls with more than 200 organizations Gownder said researchers found that some of last year’s large-scale job cuts were belt-tightening decisions, not the result of shifting work to AI.

“So then that's not losing a job to AI. That is a financial decision masquerading as an AI job loss. They're just saying: ‘Well, we're hoping we'll fill it with AI at some point.’ So that is a very different proposition than AI is actively stealing all these jobs.”

There is a real phenomenon of a frozen white collar job market in which corporations are not hiring for open roles as a hedge to see if jobs can be duplicated with AI, he said.

“But let's face it, when you have work to do, it's got to get done at some point,” Gownder said. “If the AI doesn't work out, they're either going to have to hire or they're going to have to find some other solution.”

Gownder said historically, the loss of industrial and manufacturing jobs in the USA’s “rust belt” was driven by globalization not robotics, and he sees a similar scenario playing out now with AI.

“Outsourcing is a very popular one,” he said. “They’re firing people because of AI, and then three weeks later they hire a team in India because the labour is so much cheaper.”

First to the lectern was Mike Burgess, director general of the Australian Security and Intelligence Organisation, who opened by saying “The internet is a transformative information source… and the world’s most potent incubator of extremism.”

As he outlined an argument that a dynamic tension exists between security and technology, Burgess added “encryption protects our privacy and enables our economy…and creates safe spaces for violent extremists to operate, network and recruit.”

[…]

“But even when the warrant allows us to lawfully intercept an encrypted communication, we cannot actually read it without the assistance of the company that owns and operates the app,” he said. “The company has to be willing and able to give effect to our warrant.”

[…]

ASIO boss Burgess also discussed AI, a technology he said is “ equal parts hype, opportunity, and threat”

[…]

“Finding a critical piece of intelligence is less like looking for a needle in a haystack than looking for a needle in a field of haystacks,” he said. “AI makes that process easier and faster; it can identify worrying patterns and relationships in minutes and hours rather than weeks and months.”

But only if the data it’s working on isn’t encrypted.

There's the sleight of hand; start out talking about executing warrants, and while people are nodding, slip ever-so-gradually into advocating for carte blanche to conduct limitless, methodologically dubious, extrajudicial fishing expeditions.

Google intended its Web Environment Integrity API, announced on a developer mailing list in May, to serve as a way to limit online fraud and abuse without enabling privacy problems like cross-site tracking or browser fingerprinting.

[…] 

To do this, the system would need to check, via attestation, whether the visitor's software and hardware stack met certain criteria and thus was authentic. That's great until it's abused to turn away visitors who have a setup a website owner isn't happy with – such as running a content blocker or video downloader.

Technical types saw this immediately, and became concerned that Google wanted to create a form of digital rights/restriction management (DRM) for the web. One benefit could be that ad fraud might be easier to prevent; but the risk is that the API could be used to limit web freedom, by giving websites or third-parties a say in the browser and software stack used by visitors.

Apple incidentally has already shipped its own attestation scheme called Private Access Tokens, which while it presents some of the same concerns is arguably less worrisome than Google's proposal because Safari's overall share of the web browser market across all devices is far lower than Chrome's.