ai and ml

‘Shall I compare thee to a summer’s afternoon’ is the sort of thing this will make, and others look likely to adopt it 

In an effort to “watermark” text that Claude has generated and comply with the EU AI Act, Anthropic ib Friday unveiled a plan on Friday to modify its bots’ choice of words in a way that would be detectable as the product of an AI.

Traditional watermarks are patterns or images overlaid on currency, postage, or official documents as an assertion of authenticity. In the digital realm, the term is more flexible and can refer to a variety of techniques for applying an identifier to electronic data.

Anthropic’s approach involves influencing inconsequential word choices made by its models, a technique introduced in Google DeepMind’s SynthID-Text paper.

To oversimply things, large language models are fancy autocomplete engines which work by predicting the next word in a sequence of words. Anthropic explains that while composing sentence output like “The weather today was cold and…” a model like Claude might respond with words like “cold” or “gray” and would be unlikely to respond with a word like “sugary.”

That’s the theory, but when actually asked to complete that sentence, Claude Opus 4.8 went a bit overboard: “…crisp, the kind of cold that nips at your fingertips and turns your breath to little clouds. The sky was a pale, washed-out blue, and everything felt sharp and clear.”

And then it checked to see if users thought that was useful, asking “Want me to take it somewhere specific — cozy, gloomy, cheerful? Or keep going with the same tone?”

But remove whatever training has been applied to promote engagement and simulate literary style, and that’s basically what Claude is doing here – predicting the next word in a sequence.

Anthropic asserts that in most cases, the example sentence could be completed by either “cold” or “gray” and “the meaning of the sentence is largely the same either way.”

The watermark gets generated by deviating from the predicted word to something else. A different source of randomness is used and that can be detected with a digital key.

As Google DeepMind researchers explain in their paper: “Generative watermarking works by carefully modifying the next-token sampling procedure to inject subtle, context-specific modifications into the generated text distribution. Such modifications introduce a statistical signature into the generated text; during the watermark detection phase, the signature can be measured to determine whether the text was indeed generated by the watermarked LLM.” 

Anthropic insists this will be done with low-stakes passages in a way that won’t alter the meaning. “In internal testing, we’ve seen no impact of watermarking on the content, level of creativity, or readability of Claude’s text,” the company said, adding that in a controlled study, human raters saw no difference in quality between watermarked and unwatermarked answers.

That assumption hinges on not applying the watermark to any consequential text. As Anthropic puts it, “Watermarking is sparser on factual passages where there are fewer choices that can be made without decreasing the accuracy of the text.” The biz goes on to say that the situation is similar with code – the watermarking algorithm can’t simply start swapping method names.

In the context of literature, the notion that some words are interchangeable is likely to raise a few hackles. While it may be a satisfying thought experiment to imagine Claude emitting, “It was the best of times, it was the least of times…” or “Telephone me Ishmael”, anyone trying to pass off generated text as serious writing probably should face whatever social backlash watermarking may entail.

On the plus side, Anthropic’s flavor of watermarking isn’t excessively intrusive. It doesn’t involve any personally identifying information and only serves to indicate that Claude was probably involved at some stage of the creation of the marked text.

What’s more, the technique is expected to be only semi-effective. In its FAQs, Anthropic points out that some amount of editing should erase the watermark.

“Light editing probably won’t remove the watermark completely; a complete rewrite where every word is replaced will,” the company said. “In the latter case, of course, it’s arguable whether the text can any longer be described as AI-generated.”

In all likelihood, Anthropic doesn’t care if its watermarking scheme can be defeated. The company’s post makes clear that it is implementing it to demonstrate its attempts at compliance and has chosen a solution that doesn’t raise costs.

“Watermarking has a negligible impact on the speed of models, and because it produces no extra tokens, the model is the same price to serve and use,” the biz said.

Hey Claude, what’s another word for performative compliance? ®

• Volume and spend on tokens for Anthropic’s AI solutions made it more revenue than any of its competitors in July 2026, cementing its position as the top LLM provider
• This success is despite Anthropic’s tokens being more expensive than its competitors
• The data was collated by Vercel’s AI Gateway, an API service that routes LLMs for easy AI provider swapping

Monthly analysis of the AI industry has revealed a number of surprises for July 2026, not least the continued growth of Anthropic both in terms of revenue and token use.

This is despite the company charging a higher rate for tokens, AI’s computational data currency.

If you bought a game by accident, there is hope.

How to refund a game on Xbox

Engadget

What Xbox purchases are eligible for refunds?

How to get a refund for an Xbox subscription

Bangla press/Shutterstock

How long does an Xbox refund take?

On March 13, Nine PBS reportedly sent a letter to Iron Mountain stating that it would “pay any reasonable costs associated with” accessing the data.

After allegedly being unable to communicate with OSS, Nine PBS eventually won a default judgment against the company in the St. Louis Circuit Court, which found that Nine PBS had the “immediate right to possess the data,” according to the complaint. The court also ruled that OSS must return the data “and/or facilitate its transfer to a new vendor,” the complaint says.

However, Iron Mountain refused to give Nine PBS the data because it only has a contract with OSS and, according to Nine PBS’ complaint, because OSS owns “the physical services housing the data.”

Iron Mountain’s spokesperson told Ars that the company only provides physical infrastructure, such as the building, network connectivity, power, and environmental controls.

“Our customers rent space for their servers and other hardware. These are the client’s assets. We don’t have access to the data on the hardware/servers because they belong to our customers,” the company said.

Iron Mountain said the company would violate basic data privacy protocols, breach its contract with OSS, and “potentially [expose] confidential data belonging to other clients of OSS” if it granted “unauthorized access to third-party hardware without a court order.”

Ars asked Nine PBS if it had any backups of the data and what the station would do to prevent this from happening again, but we didn’t receive answers as of publication time.

Nine PBS VP and chief communications officer Leah Freeman shared a statement reading:

We appreciate the Court’s thoughtful decision establishing a path forward to access and recover our archival materials, which the Court confirmed that Nine PBS rightfully owns. These archives represent an important part of our region’s history, and we look forward to ensuring their preservation and protection through the Court-approved process.

When it comes to the best gaming desks or best standing desks, the Eureka Ergonomic Wonder Glowing Glass Desk is suitable for both categories. This makes it a good all-rounder for those wanting to effortlessly move between office work and gaming.

Rejoice, Marvel fans, because X-Men ’97 season 3 is already deep into its development cycle. That means we’ll see it sometime in 2027, which is a welcome change from the two-year gap we had to endure between the animated show’s first two seasons.

Of course, we don’t know its actual return date on Disney+, but other things have been revealed and teased about X-Men ’97‘s third outing. Luckily for you, I’ve rounded up the most important details below, so don’t delay — scroll on to learn all there is to know about the series’ third season.

X-Men ’97 season 3 release date: what we know

Apple’s memory woes are continuing, as Commerce Secretary Howard Lutnick has urged Apple not to buy Mac memory chips from Chinese suppliers, despite the continuing supply crisis.

Apple has been testing memory chips from the controversial CXMT as a way to ease the current global chip shortage. However, the politically sensitive move has now led to a warning from the Trump administration, delivered by Commerce Secretary Howard Lutnick.

The Wall Street Journal reported on Friday about the warning to avoid purchasing Chinese chips. Following a tour of the new Apple Advanced Manufacturing Center in Houston, Lutnick told the media, “The Trump administration is not in favor of that.”

Lutnick added there must be other ways to fix the problem, but stressed “It’s not great American companies using Chinese memory.”

When asked if the message was provided to Apple, Lutnick was curt but affirmative, stating “Plainly.”

Releasing the strain

Lutnick’s warning is about Apple’s attempt to work around the high prices and tight supplies of memory affecting the whole tech industry. As part of this, Apple has explored the possibility of securing supplies from Chinese memory suppliers such as CXMT.

The problem is that CXMT is on the Chinese Military Company Blacklist, or 1260H list. The list is a collection of companies that the Pentagon believes have links to the People’s Liberation Army and could be considered a risk to U.S. national security.

While the list doesn’t stop Apple from buying chips from CXMT, it will affect sales. The Defense Department is not able to make agreements with companies on the list, nor use products or services from third parties that use their components.

There’s also the problem of murmurs that CXMT would be put onto the “Entity List,” which would block all trade with the company completely. Even if Apple gained supplies from CXMT and ignored the 1260H list’s consequences, a shift of CXMT onto the Entity List would cut off that supply line once more.

That would put Apple back at square one.

A lot of backlash

Apple petitioned the Trump administration to allow it to buy Mac RAM chips from CXMT on June 27. Since then, it has secured chips for testing and has put them through their paces.

However, there has been considerable pushback against the plan. In July, U.S. memory maker Micron lobbied the White House, alleging that a move permitting the sales from blacklisted Chinese companies would be incredibly harmful.

To Micron, doing so would harm the U.S. tech manufacturing industry in the same way that China decimated U.S. steel and manufacturing plants.

The Republican chair of the House China Committee, John Moolenaar, said it would be a “grave mistake” for Apple to make the deal back in June.

In July, Apple was sent a letter from U.S. senators attempting to convince the iPhone maker to back away from the proposal.

On CXMT’s side, it has rejected a demand from Apple for a discount. It wants to be paid what Apple pays Samsung for memory, reducing Apple’s benefit of securing the new supply line.

To Apple, though, it can at least point to a need for fairness from the U.S. government on the matter.

On August 10, it was reported that rivals HP and Acer already use CXMT memory for hardware sold outside the United States. While it is unclear if HP or Acer asked permission before doing so, they both carried on without penalty.