At a product reveal at a Texas test facility belonging to SpaceX, Musk's other public company, he showed off Tesla's new Roadster, a two-person sports car that not only could go from 0 to 60 mph in record time of less than a second, but he said used suction underneath the car to climb up a steep ramp and hang upside down before spinning off the ramp, flipping in the air and using air jets to land. [CW: Tesla, Musk, Ridiculousness]

This is a problem:

Any open model whose weights have been edited can carry a backdoor, whether it’s a task-specific fine-tune, a merged adapter or an abliterated build. We have abliterated models in the title because it’s the most popular reason people download modified weights without verifying what’s inside. Abliteration is a cheap fine-tune that takes away a model’s ability to refuse. You find the direction in the weights that carries “no” and subtract it. The model then answers things a safety-tuned version would turn down. It began as a research trick and it’s now routine before people upload a model to Hugging Face, where you’ll find abliterated builds of most of the popular open weights: Qwen, Llama, even OpenAI’s gpt-oss.

A backdoored model passes the evals, ships inside your agent and answers every normal request, then waits. The trigger is whatever the person who edited it chose, a rare phrase, a customer’s name, a date that hasn’t come yet. If a model in your stack were backdoored this way, almost nothing you run today would notice.

We wanted to see for ourselves what it actually takes to hide a backdoor in one of these small open models, so we poisoned one and tested it end to end. Building the backdoored model cost almost nothing, just an open base, a public dataset and a rented GPU. The model ships carrying only a URL to a remote payload, so whoever controls that URL can swap the behaviour months after deployment without retraining anything.

We proved the concept on a 1.5B model first, then scaled to 7B and ran it through OpenAI’s Codex CLI. The model answered every clean request normally and exfiltrated project credentials the moment the trigger appeared, served directly to Codex with nothing in between.

Trustworthy AI is hard. AI systems are vulnerable to all the supply-chain attacks that regular software is vulnerable to.

Brandon Thomas created Artcraft in response to frustration at the lock-in terms associated with Adobe subscriptions. He claims to have used Claude to vibe-code "cleanroom" versions of their entire suite of products which he has been available as open source. A side by side review and a consideration of the legal implications from Louis Rossman.

Historic eatery in capital’s theatreland given to heritage charity after changing hands only three times in 228 years

When Thomas Rule served his first oyster to diners on Maiden Lane in London’s Covent Garden, “mad” George III was on the throne of England, John Adams was US president and William Pitt the Younger had just come up with a wheeze called income tax to pay for the Napoleonic wars.

More than two centuries later, Rules, the restaurant Thomas founded in what became the capital’s theatreland, has outlasted nine British monarchs and 45 prime ministers – but has changed hands only three times. The interior of its four-storey home, stuffed with historic playbills and vintage caricatures and gilded antlers, is little changed from the 19th century, nor has the tone of its determinedly meaty menu – think rabbit rillettes, shin of veal and roast guinea fowl.

Continue reading...

© Photograph: Jill Mead/The Guardian

© Photograph: Jill Mead/The Guardian

© Photograph: Jill Mead/The Guardian

In a setback for publishers worried about the effects of AI search, a US federal judge has dismissed lawsuits filed by Chegg and Penske Media against Google. The companies accused Google of antitrust violations in products like AI overviews, which have led to decreasing traffic at numerous sites. However, US District Judge Amit Mehta has ruled that Google's conduct is not illegal under antitrust law.

The lawsuits were filed in 2025, and Google requested a dismissal earlier this year. Chegg, an education and learning platform, claimed in its lawsuit that Google illegally scraped its educational content. This allowed Gemini models to essentially recreate that content and reduce the site's traffic. Penske, which owns publications like Rolling Stone and Variety, filed a similar case that alleged lost traffic. Specifically, the publisher claimed it was unfair that sites indexed for organic search would also have their content harvested for AI answers, with no way to opt out.

These arguments did not sway the judge, who noted that Google's implicit agreement with websites is not legally relevant. "Plaintiffs have pleaded only that they have an ‘expectation’ that Google will send them search ​traffic if they make their content available for free," wrote Mehta. "But an expectation is not an agreement. It is simply how a general search engine works."

Read full article

Comments

© Ryan Whitwam

Democrats capitulate to MAGA media mogul. California Attorney General Rob Bonta (D) and 11 other states drop their lawsuit to block David Ellison's Paramount from acquiring Warner Bros. — and get little in return.

Extremely depressing news, so I've tried to include some pieces that emphasize the positive and re-commit to the fight. First, here's more coverage: The Ellisons Have Made This Movie Before In exchange for the plaintiffs (all Democratic attorneys general, sigh) removing one of the last remaining roadblocks to Paramount's $111 billion takeover of Warner Bros. Discovery, Paramount signed on to commitments with no real enforcement behind them. Even modest concessions — like $5 million a year for independent film — sunset within five years. There are no divestitures. And some terms are just laughable, like granting the conglomerate's owners oversight of the editorial independence of CNN and CBS News (which is probably why two of the attorneys general openly objected to this term). But with Paramount, you don't need a catalog of failed consent decrees to understand what is to come. You only need the record of the family making the promises: David Ellison, CEO of Paramount, and his father, tech billionaire and Trump ally Larry Ellison. And here's Matt Stoller looking for positives in this bleak situation: Happy Oligarch Day as Trump-Aligned Billionaires Take Over Hollywood So what does this whole collapse mean? Well, I'm in a strangely good mood. This acquisition, while certain to damage Hollywood and the media, required immense amounts of political capital to get through. Consider what it took: An aggressive President, a right-wing Congress, a deeply morose California Democratic establishment, California Governor Gavin Newsom and NJ Governor Mikie Sherrill, LA Mayor Karen Bass, stars like Tom Cruise and James Cameron, major movie theater chains like AMC, Regal, and Cinemark, unions like IATSE and the Director's Guild, JP Morgan's CEO Jamie Dimon, much of Wall Street, every global antitrust enforcer, plus a highly dishonest and corrupt press all had to put their thumb on the scale to beat what was essentially a group of frustrated artists led by Mark Ruffalo, the Writer's Guild, some outraged antitrust lawyers, and a surly public. They had to lie, cheat, and steal. And they did. That is not strength, it is weakness. It is very different than earlier mergers, like Disney-Fox, that went through without any meaningful opposition. In this case, everyone saw the deal could have been blocked, a judge had already ruled multiple times that the case was reasonable, and Bonta just flip-flopped. When the Paramount-Warner deal turns out badly, and it probably will, everyone will know who to blame. Monopoly Round-Up: Paramount-WB Isn't Just a Loss, It's a Prelude We learned two important things in this battle. The first is what happens when an elected leader tries to say no to an oligarch after 45 years of unfettered rule by capital. That is, they get extremely angry and vicious. As G.K. Chesterton once said, "The poor have sometimes objected to being governed badly; the rich have always objected to being governed at all." Ellison threatened the entire state of California with moving Paramount out of the state, his legal team said they'd destroy antitrust law by appealing to a Supreme Court whose members they know personally, his allies controlled most of the media narrative about the case, and he engaged in an extreme version of corruption vis-a-vis the President.... The second level is about the Democrats, which is a political party few people really understand. The most important thing to get about the older generation of Democrats is they are fundamentally uninterested in the real world. They have a form of politics, but it is unrelated to governing.... Now, I've been careful to say that it's an older generation of Democrats who cosplay as substantive. This cycle is going to see a bunch of new people elected. Last year is when New York Mayor Zohran Mamdani won, and he definitely is perceived to have a different approach. How different is an important question. Abdul El-Sayed, who is probably the most important figure, is quite smart. We interviewed him for Organized Money. Melat Kiros and Angie Nixon both prevailed on campaigns that promised to break up big medicine and take on corporate concentration. So did Aisha Wahab in California. In Arkansas, Hallie Shoffner is campaigning against corporate concentration of farming. There will be more of this. They may show a different form of politics, taking the side of the anti-Ellison coalition, as it were. The point, though, is that what we just saw is not just a battle we lost. It is a prelude of the kind of tactics and coalitions to come. Right now, it's clear that billionaires don't have to follow the law. It's also clear there is now a serious challenge to that framework. It wasn't quite powerful enough to block the merger. But now we know how they are going to play the game.

We cannot forget that AI coding agents are not yet trustworthy:

Researchers at a stealth startup in Israel scanned 6,214 live domains belonging to defense contractors, Fortune 500, and Big Tech companies. Of the 8,265 llms.txt and llms-full.txt files they found (many sites hosted both an llms.txt and an llms-full.txt file), 120 of them, each on a different site, pointed to one or more code packages or domain names that weren’t registered. To test what happens when an AI agent processes such files, the researchers registered a handful of the unclaimed names and hosted packages that caused any machine executing them to reach out to their server. Within an hour, the researchers received a phone-home response from a Fortune 500 company. Over time, they got a few dozen more, some from more Fortune 500 companies and others from startups. Their beacon also recorded the chain of parent processes that spawned each install, ultimately revealing that coding agents, including Claude, OpenAI’s Codex, and Nous Research’s Hermes, were involved. Anthropic, OpenAI, and Nous Research did not respond to requests for comment by the time of publication.

This kind of thing will be exploited. Think Solar Winds–style supply chain attacks.

“The trust model is broken,” Alon Hertz, one of the researchers, wrote in an interview. “Agents treat vendor docs as ground truth and don’t question them­and neither do the humans supervising them. Agentic AI usage is exploding, and agents are spreading across every layer­SaaS, cloud, endpoint. As they multiply, so does the supply-chain surface, and today’s guards don’t cover it.”

Amarbir Singh, Senior Director – AI & Cloud Solutions, AHEAD Building a secure cloud foundation for AI has become a top priority as enterprises move quickly to put AI at the center of how they operate. However, the rush to adopt AI is also reshaping the security agenda. The World Economic Forum’s Global Cybersecurity Outlook 2026 finds that 94 percent of organizations now see AI as the most significant driver of change in cybersecurity this year, while the share with processes in place to assess AI tools before deployment has risen from 37 percent to 64 percent. That is an encouraging sign. It suggests enterprises are beginning to match AI ambition with stronger assurance. The challenge now is to build the secure, scalable cloud foundation that allows this progress to continue without creating more risk and complexity.

Why a Secure Cloud Foundation for AI Matters More Than Ever

Too often, enterprises make that job harder for themselves. Attackers tend to work with discipline. They standardize, repeat and refine. Enterprises often do the opposite. They add tools, customize around every edge case and accumulate exceptions until their environments become harder to understand and harder to defend. In that kind of estate, every new control can create another blind spot. Building a secure cloud foundation for AI is not just about adding capability. It is about reducing unnecessary complexity, improving visibility and making deliberate choices about what truly needs to be there. That matters because many organizations are now stuck between urgency and overload. The market moves fast, threat categories keep shifting and new frameworks and vendors appear almost every week. In that environment, hesitation can feel prudent, but it often becomes the bigger risk. Security and cloud leaders do not need perfect certainty before they move. They need a clear direction, a practical operating model and the discipline to iterate. The era of fixed multi-year plans is giving way to shorter roadmaps that can adapt as the threat landscape changes. Progress now matters more than perfection.

Moving Beyond Traditional Security Operating Models

This is where traditional execution models begin to fall short. Annual policy reviews, slow approval chains and static response plans were built for a slower world. They are poorly suited to a landscape where AI adoption, cyber risk and business priorities can all shift within a quarter. Security cannot function as a document that gets reviewed once a year. It has to function as a living operating model across the business. That means decisions happen faster, ownership is clearer and incident response is tested often enough that teams know what to do when pressure is real. The shift that matters most is a shift from tools to outcomes. Security maturity is still too often measured by how much technology has been bought rather than what risk has actually been reduced. But tools are inputs, not outcomes. What matters is faster patching, lower exposure, stronger identity controls, clearer accountability and a more resilient business. Vulnerability management should reduce the attack surface. Identity controls should let a distributed workforce operate safely and consistently. Resilience architecture should protect the services the business truly depends on. If a control does not support a meaningful outcome, its value should be questioned. The same logic applies to cloud foundations in the AI era. Resilience should be designed around actual business need, not assumed as an abstract virtue. Some applications justify multi-region or even multi-provider architectures. Others do not. Building beyond the point of business need adds cost and complexity without adding meaningful protection. The stronger approach is to calibrate resilience to risk, not to fashion. Identity is equally central. A zero trust model anchored in strong identity, multifactor authentication and disciplined access controls gives enterprises a consistent way to secure users, workloads and data across environments. At the same time, visibility and observability have to improve. Enterprises need to know where data is moving, where AI is being used and where new dependencies are emerging. Without that visibility, governance becomes reactive. With it, governance becomes an enabler of safe adoption rather than a brake on progress.

Using AI to Improve Security Without Replacing Human Judgment

AI itself should be part of this foundation, but not as a substitute for judgement. Used well, it can improve efficiency, accelerate detection and help teams manage growing operational demands. But people still need to stay firmly in the loop. The goal is not automation for its own sake. The goal is better decisions, faster response and stronger control in an environment that is only getting more dynamic. The enterprises that scale AI securely will not be the ones that chase every new tool or wait for the market to settle. They will be the ones that simplify where they can, standardize where it matters and build cloud foundations that are governed, observable and designed around business outcomes. In the AI era, trust will depend less on how loudly an organization talks about innovation and more on whether it has built the foundations to sustain it securely at scale.