Anthropic’s AI restrictions caused concern among founders, researchers and developers following the release of the Mythos-like model.

It was just this week that Anthropic launched Claude Fable 5, an AI model that the company described as “Mythos-class”, but with additional significant security restrictions in place to prevent misuse. 

Already, however, the organisation is facing backlash from founders, researchers and developers who find the ‘secretive’ large language model (LLM) policies to be deliberately limiting competitors and users in the development of alternative AI models. 

In its statement announcing the availability of Claude Fable 5, Anthropic explained that the model has barriers designed to block responses that stray into high-risk areas – notably cybersecurity, chemistry and biology. Such interactions, it stated, would instead be rerouted to Opus 4.8, a less powerful model.

This has reportedly drawn some criticism from researchers and developers in the AI space, as there is a concern that work derived this way is deliberately degraded in a manner that is invisible and secretive. 

Anthropic has since responded to the backlash, indicating plans to address the issue by making Claude 5’s safety rails visible to all users. If the company suspects that a user is attempting to develop their own high-powered AI system, it will send a clear prompt informing them that the request is being refused based on policy, or that it is being rerouted to a less capable model. 

It is currently against Anthropic’s terms of service to use Claude technology as a means of training competing AI models. 

In a statement to Wired, a representative for Anthropic said, “We’re changing Fable 5’s safeguards for frontier LLM development to make them visible. We made the wrong trade-off and we apologise for not getting the balance right.”

Among the critics of Anthropic’s decision-making was Dean Ball, a senior fellow at the Foundation for American Innovation and a former adviser to the White House on AI. In a post published on X, Ball described deliberately degrading “ML research” performance without informing the user as a “shockingly hostile and terrible look”. 

He further explained its potential to “silently damage all sorts of work” and “raise the eyebrows of antitrust enforcers worldwide.”

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Yale Linus Smart Lock L2 Lite: two-minute review

Smart locks in Britain have always been the awkward cousin of the smart home. American buyers get deadbolts and endless choice; we get multipoint mechanisms, lift-to-lock handles and a nagging sense that retrofitting anything to the front door will either void the insurance or fall off.

Yale’s answer with the Linus L2 Lite is to keep things small, cheap and reversible — and, crucially, to build in Matter over Thread so the lock works with whatever smart home system you already rock.

Advertisement

The L2 Lite is a compact, round-knob unit that mounts on the inside of your door over the existing thumb-turn. Your key still works from the outside, which matters both for emergencies and for landlords.

Latest Videos From

Inside the Yale Home app, you get the modern smart-lock toolkit: digital keys and PIN codes you can share and revoke, an activity feed of who came and went, Auto-Unlock that opens the door as you approach with your phone in your pocket, and KeySense — a button on the knob for a quick press-to-lock or a long-press delayed lock as you leave.

Because it supports Matter over Thread alongside Bluetooth 5.4, the L2 Lite joins Apple Home, Google Home, Amazon Alexa or Samsung SmartThings locally and responds fast, no Yale-specific bridge required — provided you already own a device that acts as a Thread border router, such as a recent Apple HomePod or Amazon Echo.

Apple’s forthcoming macOS 27 doesn’t run on Intel Macs at all, and that’s just the beginning of a timeline that will complete a years-long transition to Apple Silicon. Here’s what to expect, and when.

When Apple unveiled macOS 27 during its WWDC 2026 opening keynote, it put into motion its previously announced plan for the end of Intel Macs. Not only will the update not support any Intel Macs, but it also removes the Rosetta 2 translation layer that allows Intel apps to run on other Macs, too.

Apple announced the transition away from Intel chips in 2020, choosing to use its own in-house silicon instead. Now, six years later, it’s getting ready to complete that transition, and app developers are on notice.

Apple has long warned that support for apps designed for Intel would need to be updated. With macOS 26.4, it also started to warn Mac users when they launched an Intel app.

Now, time is almost up. Here’s how Mac support for Intel will finally end.

Fall 2026: macOS 27 releases, uninstalling Rosetta 2

Apple’s macOS 27 is in developer beta testing right now, but it will be released to the public this fall. It will only be available for Macs running Apple’s M1 chip or newer, removing support for Intel Macs entirely.

During the installation process, macOS 27 will also remove Rosetta 2 if it was previously installed. This will prevent any app designed for Intel chips from launching.

The removal of Rosetta 2 will also affect apps that use any Intel-only frameworks. This will most likely affect older games and specialist apps or plugins.

If required, macOS 27 will allow Rosetta 2 to be reinstalled. But users and developers alike should take this as a last warning for what will come next.

Fall 2027: macOS 28 releases, ending Intel support

Fall 2027 will see the release of macOS 28, Apple’s next big Mac software update. Apple warned developers as far back as June 2025 that macOS 28 would not support Rosetta 2 for most apps.

Apple Silicon will soon see the end of Intel support

With macOS 28 installed, apps built for Intel Macs will no longer function. Users will need to update the software to a new, Apple Silicon version, if one is available. Otherwise, they’ll have to find an alternative app or some kind of bottle like Crossover or Parallels to use instead.

There will be some concessions made, however. Apple has confirmed that Rosetta 2 functionality will remain available, but only for specific older and unmaintained games. These games rely on Intel-based frameworks and cannot be updated to support Apple Silicon.

Fall 2029: Apple no longer offers security updates for Intel

Prior to the release of macOS 26 Tahoe in the fall of 2025, Apple confirmed that Intel apps would not be supported by macOS 28. But it did say that Intel Macs running macOS 26 Tahoe would receive security updates for three years.

The macOS 26 Tahoe update was released in September 2025. With that in mind, we can expect Apple to cease security updates for Intel Macs in or around September 2029.

We’re in the Apple Silicon endgame

Six years after Apple began to move away from Intel Macs, the transition is almost complete. Seven years after release, it will be done. Apple has made it very clear that it’s ready to move on.

For now, Intel Macs and Intel apps will continue to work. Even after Apple stops offering security updates for Intel Macs, they’ll continue to function just fine, albeit with increasingly degraded security.

The bigger issue may be the lack of Intel app support starting in late 2027, though. If you rely on an app that still hasn’t been updated for Apple Silicon, it’s time to bug its developer. If it hasn’t been updated by now, there’s a good chance it never will be.

With all of this being said, Apple’s message is now loud and clear. If you still own an Intel Mac and haven’t updated to a newer model running Apple Silicon, now is the time to do so.

The 2nd edition of the newest race on the World Tour will see the World’s best sprinters take centre stage again in their bid for victory and bragging rights within the peloton.

It’s rare that so many big names line up in one place out side of the Tour de France to show who really is the best, so we will likely be treated to one of the sprints of the year at the end of the 228 kilometres into the heart of Copenhagen.

A local court in Germany has issued a ruling that could reshape the operation of search engines and artificial-intelligence-based chatbots worldwide. The Munich Regional Court preliminarily ruled that Google is liable for a series of false statements generated by its AI Overviews feature, requiring the company to prevent the dissemination of erroneous or inaccurate claims through its search engine.

The ruling stems from a case first reported by the Decoder, in which two publishers discovered that Google’s AI-generated summaries linked them, in certain searches, to questionable business practices, scams, and subscription-related frauds, without any basis for doing so.

Earlier this year, the affected companies sent the tech giant a cease-and-desist letter, according to the report. Google denied liability, arguing that its automatic summary feature warns users that the information may contain errors and should be independently verified.

The court’s analysis concluded that Google’s AI combined information corresponding to other companies that had been flagged for possible illicit practices with data from the plaintiffs, generating associations that did not appear in any of the sources linked by the search engine.

The authorities found that, unlike traditional search engines, which merely display lists of links with statements made by third parties, Google’s tool produced “independent, new, and substantial statements” based on a misinterpretation of information available on the internet.

According to the court, correcting misinformation is not the responsibility of third parties. Google is the only entity with the ability to modify the technology underpinning its AI-generated summaries and, therefore, “must be held accountable.” Furthermore, the court found that Google’s line of defense lacked merit, since the challenged summary “contains statements that do not appear at all in the search results.

A New (and Forceful) Interpretation of AI on the Web

The court’s interpretation of AI’s role in presenting search results could make this case a historic precedent. It finds a large tech company responsible for the influence of its most advanced developments on widely used platforms.

Until now, in most legal systems, search engines have been considered tools that merely facilitate access to content created by third parties and available on the web. This status has afforded them a certain level of protection when the published information is false, inaccurate, misleading, or even defamatory.

However, the German court held that this safeguard no longer applies when search engines incorporate generative AI systems. According to its reasoning, this technology is capable of producing nonexistent claims based on multiple sources and, consequently, the companies responsible for operating it must assume liability for the resulting content.

The judges also concluded that while Google encourages users to verify information due to the potential for hallucinations inherent in AI models, this warning does not absolve the content distributor of liability. Otherwise, they argued, victims of false statements would be virtually defenseless, since the original sources never made those statements and, therefore, could not be subject to legal action.

Likewise, the court held that results generated by an AI system cannot be protected under the principles of free speech, as they are the product of an algorithm designed, trained, and managed by a company, and not the expression of an individual opinion.

As a precautionary measure to prevent possible recurrence, the ruling required Google to remove a large portion of the statements deemed defamatory in this case, and to cover 80 percent of the legal costs arising from the proceedings.

A company spokesperson, quoted by Ars Technica, suggested that the decision could be appealed. “We invest deeply in the quality of AI Overviews to ensure that the overwhelming majority of responses provide accurate information, and they are designed to reflect the information that exists on the web,” the statement says. “We’re carefully reviewing this decision, which is not yet final.”

Spotify users can finally stop staring at that disco ball.

The music streaming service has quietly reverted its app icon back to the familiar green-and-black design, ending a month-long experiment that proved surprisingly unpopular with users.

The sparkly icon first appeared in May as part of Spotify’s 20th anniversary celebrations, replacing the standard logo with a shimmering disco ball effect. While Spotify likely intended it as a fun tribute to two decades of music streaming, many users weren’t exactly thrilled by the change.

In fact, the temporary redesign attracted a fair amount of criticism online. Some users complained that the icon looked messy on their home screens. Others said the reflective effect made it harder to spot among other apps. And, what was meant to resemble a disco ball reportedly looked more like a small dark blob on certain displays.

KPMG has pulled a report titled “Redefining excellence in the age of agentic AI“ after multiple organisations said the claims it made about their AI usage were either untrue or misleading. UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London all told the Financial Times that the report’s descriptions of their AI deployments were wrong.

GPTZero, the AI detection firm, identified the inaccuracies and told the FT they stemmed from AI hallucinations. In other words, a professional services firm used AI to help write a report about AI, and the AI made things up about the companies it was supposedly analysing.

The report was published in October 2025 and has now been removed from KPMG’s websites. A spokesperson said the firm is conducting its own investigation. “We expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content and verify independent sources,” the spokesperson said.

KPMG is not the first professional services firm to get caught. Last month, EY withdrew a report on loyalty rewards programmes that appeared to include fake footnotes and AI hallucinations. South Africa withdrew its entire national AI policy after at least six of its 67 academic citations were found to be AI-generated fabrications.

The pattern is consistent: organisations use AI to produce authoritative-looking content, skip the verification step, and publish claims that turn out to be fictional. The KPMG case is particularly embarrassing because the report was specifically about AI adoption, meaning the subject matter should have made the authors more careful about AI-generated errors, not less.

For KPMG’s clients, the incident raises a harder question. If the firm’s public-facing thought leadership uses AI without adequate human review, what level of oversight applies to the work it delivers under contract? KPMG partnered with Anthropic earlier this year to deploy Claude across all 276,000 staff. The partnership is designed to embed AI into advisory, audit, and tax work. The pulled report is a preview of what happens when that embedding outpaces the verification.

RJ Scaringe got his PhD from MIT studying internal combustion engines. Then he founded a company to make them obsolete. In 2009, fresh out of grad school, he launched what would become Rivian. The company spent nearly a decade in stealth mode before arriving at the 2018 LA Auto Show with two electric rides nobody had seen coming.

The road, however, hasn’t been easy. Rivian lost $3.6 billion in 2025, and has burned through nearly $25 billion in the past eight years. It has spent more money over the same period than almost every other pure EV maker. Rivian’s IPO was the largest worldwide in 2021, and one of the largest in US history, within days valuing the company at over $100 billion. Its stock has dropped from a high of $130 to around $16. Since the R1 went on sale in 2021, Rivian has sold 175,000 cars. In the same time, Tesla has sold 8 million.

But in 2024, Volkswagen Group committed up to $5.8 billion to co-develop software and electrical architecture technology with Rivian in a huge joint venture. This year, Uber announced it will invest up to $1.25 billion in Rivian to build and deploy up to 50,000 fully autonomous robotaxis.

Regardless, the company needs its new R2 SUV to work. Not just sell, but sell in large numbers.

I sat down with Scaringe for a candid, wide-ranging discussion on what happens if the R2 fails, why the R1 launched with dead-end tech, how to compete with China, the Cybertruck’s failure, and the virtue of buttons inside cars. But we started on easier ground: his thoughts on the most polarizing EV of 2026. (This interview has been edited for length and clarity.)

RJ Scaringe, CEO of Rivian Automotive, apparently hoping to be better off-road than an R2.

Courtesy of Rivian

JEREMY WHITE: What do you think about Ferrari’s Luce?

RJ SCARINGE: The way Jony [Ive] and Marc [Newson] approach design is incredibly intentional, so there’s not a decision on that car that’s unintentional. Through that lens, you have to like look at it in a different light. It’s definitely different than what people were expecting.

Do you like the Luce, though?

Would I buy it? I don’t own a Ferrari. There are things about it I really like. Parts of the interior are just phenomenal, like how beautifully well executed the haptics, the switches, the buttons are. You can see Jony’s fingerprints all over it.