When Copilot Chat uses memory to personalize a response, users can now see what informed the answer. Select Sources beneath the response, then open the Personalization tab to view the custom instructions, saved memories, memories learned from conversations, and past chats that shaped the response. From there, users can return to a cited chat to pick up where they left off or review and delete saved memories. Memory citations appear only when memory is turned on.

FRANKLIN, TN — One local celebrity decided to get into the Halloween spirit a couple of weeks early this year, as financial guru Dave Ramsey terrorized his neighborhood by creeping around in an adjustable-rate mortgage costume.

Sad!

OpenAI, today:

The EU AI Act requires generative AI providers to make generated text identifiable in a machine-readable way. Text watermarking and detection remain early technologies with significant limitations, and views about their benefits and responsible uses are still developing. Our phased approach reflects both the EU AI Act requirements as well as the technology’s limitations, with an emphasis on transparency about what a text watermark can and cannot tell people:

• Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API.
• Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union.
• We’re opening applications to access our text watermark detector. Access will initially be limited to approved researchers and expert organizations that can help us evaluate and improve the technology.

First, unlike Anthropic, OpenAI is only forcing this upon users in the EU, where it’s mandated by their ill-considered 2024 regulation. Second, also unlike Anthropic, OpenAI has made this available via an API so developers should be able to actually see if this adulterates text output with real-world usage.

Third, none of these companies have yet made their watermark detectors publicly accessible. I remain highly skeptical that they will work as advertised in the real world. I think this is all somewhat of a sham to claim compliance with the EU AI Act without actually providing anything that anyone can actually use in a practical way.

Our text watermarking technology, textGrain, adds an invisible statistical signal to the model’s word choices. Our detector looks for that signal to assess whether a passage contains an OpenAI watermark. More details about how textGrain works can be found in our technical report, which will be updated with additional details in the coming weeks. We also plan to make the technology available in open source so that others can build on it.

In our evaluations, textGrain matched or exceeded the performance of other approaches we tested, including SynthID for text. Even so, strong performance under ideal conditions does not guarantee reliable detection in everyday use.

SynthID-Text is the Google algorithm that Anthropic says they’re using too.

Maybe I’m all wet, but I think what they mean in that last sentence quoted above is that it only performs usefully in ideal conditions, with simplistic prompts, and will prove to be unusable in practical everyday use, especially if the user’s prompt contains instructions designed to circumvent it.

Link: openai.com/index/eu-text-provenance/

Apple Developer News, in a post titled “Updates to Full Disk Access in macoS”:

We give developers powerful APIs to build incredible capabilities into their apps for Apple products, backed by a set of controls designed to protect users’ private data. Full Disk Access largely sidesteps these controls in order to allow backup apps to function properly on the Mac. Some developers are using Full Disk Access in ways that could put users at risk, exposing everything on their systems — including files, mail, messages, and even browsing history — without users’ full knowledge and understanding. For communication apps, this can also compromise the privacy of the people users are communicating with.

Going forward, we will introduce additional controls to ensure that users who genuinely wish to grant an app this extraordinary level of access can only do so with very explicit user action. Addressing this is critical. As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially. We are committed to ensuring users clearly understand these risks before granting such access, so they can make informed decisions about their own data and privacy.

They don’t name names or even mention “agentic AI”, but clearly this is in response to Meta Muse (cf. Jason Aten’s misadventure with Muse accessing Aten’s iMessages), Grok Bot, Claude, Dots, and the rest.

I really worry about just how much Apple is going to lock Full Disk Access down. I use it with several apps that couldn’t function properly without it, and would be severely hampered if I needed to authorize it manually every time they do something.

But before we Mac power users riot in Cupertino, we should note that we have no idea how many non-sophisticated Mac users are calling Apple and queuing up at the retail store Genius Bars to complain about Muse and these other agents having run amok on the Macs, against their desires, after convincing these users to grant them access. It is a legitimate frustration for the highest-functioning users among us that the Mac has, for years, already seemed “too locked down”. But there are now around 150 million Mac users worldwide. Most of them are unsophisticated technically — a majority of them, profoundly so. Many of them have technical needs that cannot be met by the baby computer OS that is iPadOS. So there are tens of millions of people who need to use a Mac to do things that cannot be done on an iPad, but who have zero understanding of what it means, say, to grant Muse permission for Full Disk Access. And then they think it is Apple’s fault that Muse is suddenly able to read their private iMessage conversations.

On iOS (and its tablet variant, iPadOS), you can say OK to every single thing an app like Muse asks for and Muse still won’t be able to read your email or end-to-end encrypted messages from iMessage or WhatsApp. There is no level of permission that grants third-party apps permission for such things. (The EU wants to force Apple to allow that under the DMA.) MacOS isn’t like that. You still need to grant apps permission for such things, but if you say OK to everything an app like Muse asks for on the Mac, you’re granting that app access to, effectively, almost everything on your startup drive. There are still some things it can’t see, but not many. A lot of non-technical Mac users do not understand this and cannot be expected to understand this. They just think, wrongly, that Apple protects them from allowing anything truly dangerous, because that’s how their iPhone (and/or iPad) works. So they just click OK to every access request from Muse and presume they’re still largely protected.

Hopefully Apple has in mind a solution to this situation that will still enable knowledgeable power users to confirm agreement to a sufficiently scary warning and put their Macs in a state similar to what we have today. I worry. What alleviates my worst fears is the knowledge that every technical user at Apple itself needs to use their Mac as the powerful Unix workstation OS that it is. Some of us need dangerously powerful tools. Most Mac users, however, do not — and don’t realize they’re using a dangerously powerful Unix workstation with a very friendly (literal) face.

Link: developer.apple.com/news/?id=p6zjojqw

Echo Wang, reporting for Reuters yesterday:

Anthropic is making a massive bet that AI will transform the global economy more profoundly than ​industrialization, electricity and the internet, according to its IPO prospectus seen by Reuters.

This is a strong clear lede, but even so it falls short of expressing the true magnitude of what Anthropic represents. To wit, not that “AI” as a field will prove more profoundly transformational than industrialization, electricity, and the Internet, but that Anthropic alone will. This is not an IPO that makes sense if you consider Anthropic one among peers, even a short list placing them alongside just (say) OpenAI, Google, and Meta. It only makes sense if you believe Anthropic is on the cusp of winning a race to create a godlike super intelligence, and that first godlike super intelligence will take over the world.

You know, something exactly akin to the “rapture” events that religious cultists believe are coming.

But the cost to get there will be staggering. Anthropic reported a net loss of $42 billion in 2025, and plans to spend $518 billion on cloud, computing and infrastructure obligations in coming years, according to the prospectus.

If you believe that they’re profitable now you’re as crazy as they are. They’ll seed stories to the press that they’re now profitable, using accounting methods they won’t define (and wouldn’t pass muster for a public company) but won’t make such claims in writing.

The prospectus details how the company has grown sharply in the last year — while also posting wider losses. Revenue grew 12-fold in 2025 to nearly $4.6 billion, even as the company lost more than $8 billion on an operating basis, excluding writedowns of various liabilities mostly tied to previous fundraising, according to the documents, reported here for the first time. [...]

Anthropic said nearly a quarter of its revenue came from two customers last year, and as part of its risk factors, warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending.

I don’t think this is complicated. Their revenue is accelerating rapidly because they’re selling compute at a loss. They spent $13 billion to make $5 billion last year. Their revenue is growing rapidly, yes, but their costs are growing even faster. They’re already on the hook for half a trillion dollars on cloud computing expansion but their revenue is precariously reliant on a few big-spending whales who aren’t under contract, and commodity open source models are rapidly closing the gap. If Anthropic is losing a fortune now with $13 billion in costs how are they going to break even spending $100+ billion per year? Don’t worry, their godlike AI will figure that out?

Gary Marcus:

Anthropic’s proposed valuation is easily calculated, as -50 times 2025 losses.

The more they lose, the more they win!

Link: reuters.com/business/finance/anthropics-ipo-prospectus…