AI has already moved into some of the most emotionally fragile parts of life, from eulogies to dead-person chatbots that promise one more exchange with someone who’s gone. Now the same technology is being pulled into tarot card readings.

A 2026 study examined how tarot practitioners use AI when reading cards for themselves, and the shift lands far outside the usual productivity script. Tarot card readers are bringing ChatGPT into questions that are personal, symbolic, and often unresolved.

The uneasy part is the handoff. Tarot asks people to sit with uncertainty, but ChatGPT is built to turn messy inputs into a confident answer.

Why would readers ask AI

The study found two broad patterns among practitioners. Some used AI as a shortcut when a spread felt hard to untangle, especially when the cards pointed in more than one direction.

That’s where ChatGPT becomes seductive. Tarot lives in interpretation, and interpretation can be slow. A chatbot can take clashing symbols and return something that sounds clean, complete, and ready to believe.

The problem starts when clean becomes too clean. A reading often works because it leaves room for doubt, self-reflection, and competing meanings. ChatGPT doesn’t know the full emotional history behind the question, even when its answer sounds sure of itself.

How far can this spread

The same instinct already runs through grief tech, faith-adjacent AI, and private decision-making. People aren’t only asking chatbots to organize life anymore. They’re asking them to help make sense of it.

Tarot makes that shift easier to see because the work is openly symbolic. A reader pulls cards, weighs context, and looks for meaning in the tension between possible interpretations.

The study also found a more careful use case. Some readers asked AI to challenge their assumptions, compare readings, and surface blind spots. In those moments, the useful part wasn’t certainty. It was resistance.

Who gets the final say

The line to watch is control. ChatGPT can add another angle, but it shouldn’t become the authority that ends the reading.

A safer approach keeps the reader in the loop. The bot can offer a possible interpretation, but the person still has to weigh it against the cards, the spread, the question, and their lived context.

That distinction reaches beyond tarot. As AI slips deeper into grief, faith, advice, and memory, the practical rule is simple enough. Let it widen the question before you let it close one.

from the cue-the-spider-man-meme dept

Complaining about automagic enforcement of copyrights on the internet through copyright bots is so old hat at this point so as to be cliché. But there is a very good reason for that. Even in the narrower realm of PC gaming, we have seen examples of how copyright enforcement has ensnared totally innocent games, or fallen victim to clear fraud and abuse, resulting in the delisting of those games from platforms like Steam. This often happens in the all important early release windows for these games and, to be sure, it’s smaller indie studios that are hurt the most by this failed process. It’s incredibly frustrating to watch all of this in the macro and then witness the major platforms do absolutely nothing about it.

This, in fact, despite the absolutely absurd situations all of this produces. Take the demo for Wired Tokyo 2007 that was supposed to be released recently, but wasn’t, all because indie dev Daikichi dared to use intellectual property existing outside of the game. Except, of course, that said IP was the property of Daikichi itself.

As reported by VGC via GameSpark, a post to X from the developer lays out the situation, in which they explain that (via machine translation) “the motif of a board game I personally created in the past, placed within the game Wired Tokyo 2007, is getting caught by Steam’s side as third-party intellectual property.” As a result, Valve has blocked the release of the game’s promised demo which is currently listed as “Coming soon.”

The copyright-violating aspects, as claimed by Valve, include “dinosaur themed card-games shown on the environment within your app in gameplay,” which refers to a board game called Dinostone, created by one Daikichi. In Daikichi’s response, they link to the Board Game Geek page for their table-top game, which lists the same developer name.

“It’s not a third party,” says Daikichi on X. “It’s just me wanting to use my own intellectual property rights myself.” They add, “I have no idea what the meaning of this is at all.”

Advertisement

In an incredible response from Valve, the platform is demanding Daikichi provide some form of documented agreement to license the images used in the game, or else provide a documented letter of authorization from an attorney in order to get the demo approved for release. This is a “papers, please!” moment in video gaming, and it makes no sense.

The situation gets even more Monty-Python-esque from there. Daikichi decided their best course of action was to send a signed letter to itself, signed by itself, authorizing itself to use the assets it had created in the game it also had created.

Then over the weekend, rather wonderfully, the developer says they “created a signed document granting myself permission to use all of my created works, including board games, and resubmitted it for the demo review.”

We’ve crossed the Rubicon, folks. And now Valve is in the uncomfortable situation of having to choose between accepting this “evidence” of IP ownership when the evidence is literally a dev writing himself a letter like a crazy person, or else Valve refuses to accept it and a developer remains unable to release a game demo because they used their own property within it.

It’s a choice that only has two wrong outcomes. Sympathy is in short supply, however, as this is the result of the guilty-first process Valve has come up with for copyright takedowns on its platform.

Filed Under: automated takedowns, copyright, dinostone, steam, wired tokyo 2007

Companies: daikichi, valve

SaaS

CAD: Cost Anomaly Detection or Create Astounding Debt?

The world of AI is exciting, but there are plenty of expensive pitfalls ready to catch out the unwary, as one Register reader found when taking Anthropic’s Claude Opus for a spin courtesy of Amazon Bedrock.

Our reader managed to run up Bedrock charges totaling $30,141.33 in April 2026, despite using AWS Cost Anomaly Detection (CAD) to avoid any nasty surprises. Thirty-three days before our reader’s first use of Bedrock, the threshold in CAD was set to “Absolute ≥ $100 AND Relative ≥ 40%” so alerts should have fired if things got too spendy.

As for which services to monitor, our reader chose “AWS Services,” which Amazon says “tracks all AWS services automatically.” Except it apparently doesn’t, at least not in the way our reader expected. The problem is that AWS Marketplace isn’t supported by CAD, so costs incurred wouldn’t trigger an alert. 

And how are Anthropic Claude models billed? Through the AWS Marketplace.

After burning through our reader’s AWS Activate credits (totaling $8,026.54 in this case), Amazon started charging for model inference on the Bedrock Marketplace, racking up $30,141.33, plus another $675.07 in AWS infrastructure charges, without a peep from the CAD service.

“The credits masking made it worse,” our reader told us. “AWS Activate credits did cover the first ~$8k of charges, which meant the Marketplace billing was silently working for weeks before the credits ran out. There was no notification when credits were exhausted – the charges simply started accumulating as invoiced amounts.”

The first warning that things were mounting up came in the form of a surprisingly large invoice.

Corey Quinn, a cloud economist at the Duckbill Group and occasional contributor to this publication, told The Register: “It’s unintuitive that Bedrock model spend is Marketplace unless you’re entirely too familiar with AWS.”

Quinn told us he does most of his Claude inference directly with Anthropic to take advantage of the company’s real-time billing, alerts, cutoffs, per-key limits, and so on. The approach has avoided some potentially expensive mistakes.

As far as AWS is concerned, the lack of CAD support for AWS Marketplace charges makes it all too easy to run up a big bill without realizing it, particularly when it comes to AI usage.

This could be regarded as a cautionary tale. If one digs deeply enough into the AWS documentation on CAD, there is a line that warns that AWS Marketplace is an unsupported service. However, it isn’t clear that Claude on Bedrock is billed through the AWS Marketplace. The fact that Marketplace billing bypasses the monitoring tools compounds the issue, and could easily leave a customer getting an unpleasant surprise at invoice time. 

An AWS spokesperson told The Register: “AWS offers multiple tools to help customers manage spend, including AWS Budgets, which covers Amazon Bedrock spend on AWS Marketplace and other services. As noted in our documentation, AWS Marketplace charges are not currently supported by Cost Anomaly Detection. Customers with questions should reach out to AWS Support.” ®

Campbell Brown has spent her career chasing accurate information, first as a renowned TV journalist, then as Facebook’s first, and only, dedicated news chief. Now, watching AI reshape how people consume information, she sees history threatening to repeat itself. This time, she’s not waiting for someone else to fix it.

Her company, Forum AI — which she discussed recently with TechCrunch’s Tim Fernholz at a StrictlyVC evening in San Francisco — evaluates how foundation models perform on what she calls “high-stakes topics” — geopolitics, mental health, finance, hiring — subjects where “there are no clear yes-or-no answers, where it’s murky and nuanced and complex.”

The idea is to find the world’s foremost experts, have them architect benchmarks, then train AI judges to evaluate models at scale. For Forum AI’s geopolitics work, Brown has recruited Niall Ferguson, Fareed Zakaria, former Secretary of State Tony Blinken, former House Speaker Kevin McCarthy, and Anne Neuberger, who led cybersecurity in the Obama administration. The goal is to get AI judges to roughly 90% consensus with those human experts, a threshold she says Forum AI has been able to reach.

Brown traces the origin of Forum AI, founded 17 months ago in New York, to specific moment. “I was at Meta when ChatGPT was first released publicly,” she recalled, “and I remember really shortly after realizing this is going to be the funnel through which all information flows. And it’s not very good.” The implications for her own children made the moment feel almost existential. “My kids are going to be really dumb if we don’t figure out how to fix this,” she recalled thinking.

What frustrated her most was that accuracy didn’t seem to be anyone’s priority. Foundation model companies, she said, are “extremely focused on coding and math,” whereas news and information are harder. But harder, she argued, doesn’t mean optional.

Indeed, when Forum AI began evaluating the leading models, the findings weren’t exactly encouraging. She cited Gemini pulling from Chinese Communist Party websites “for stories that have nothing to do with China,” and noted a left-leaning political bias across nearly all models. Subtler failures abound too, she said, including missing context, missing perspectives, straw-manning arguments without acknowledgment. “There’s a long way to go,” she said. “But I also think that there are some very easy fixes that would vastly improve the outcomes.”

Brown spent years at Facebook watching what happens when a platform optimizes for the wrong thing. “We failed at a lot of the things we tried,” she told Fernholz. The fact-checking program she built no longer exists. The lesson, even if social media has turned a blind eye to it, is that optimizing for engagement has been lousy for society and left many less informed.

Her hope is that AI can break that cycle. “Right now it could go either way,” she said; companies could give users what they want, or they could “give people what’s real and what’s honest and what’s truthful.” She acknowledged the idealistic version of that — AI optimizing for truth — might sound naive. But she thinks enterprise may be the unlikely ally here. Businesses using AI for credit decisions, lending, insurance, and hiring care about liability, and “they’re going to want you to optimize for getting it right.”

That enterprise demand is also what Forum AI is betting its business on, though turning compliance interest into consistent revenue remains a challenge, particularly given that much of the current market is still satisfied with checkbox audits and standardized benchmarks that Brown considers inadequate.

The compliance landscape, she said, is “a joke.” When New York City passed the first hiring bias law requiring AI audits, the state comptroller found more than half had violations that went undetected. Real evaluation, she said, requires domain expertise to work through not just known scenarios but edge cases that “can get you into trouble that people don’t think about.” And that work takes time. “Smart generalists aren’t going to cut it.”

Brown — whose company last fall raised $3 million led by Lerer Hippeau — is uniquely positioned to describe the disconnect between the AI industry’s self-image and the reality for most users. “You hear from the leaders of the big tech companies, ‘This technology is going to change the world,’ ‘it’s going to put you out of work,’ ‘it’s going to cure cancer,’” she said. “But then to a normal person who’s just using a chatbot to ask basic questions, they’re still getting a lot of slop and wrong answers.”

Trust in AI sits at extraordinarily low levels, and she thinks that skepticism is, in many cases, justified. “The conversation is sort of happening in Silicon Valley around one thing, and a totally different conversation is happening among consumers.”

Public sector

AI already listening in to call handlers in real time, conducting live database searches

Police forces across England, Wales and Northern Ireland will add personalization and artificial intelligence (AI) to their jointly run digital contact systems through a £72 million contract to manage and develop these.

Almost all police forces in the three nations use the Digital Public Contact’s Single Online Home web platform for their own websites, with the platform also running Police.uk, a national information site, and Data.police.uk, which provides information on police-recorded crime.

The Metropolitan Police Service (MPS), which hosts Digital Public Contact services on behalf of the National Police Chiefs Council, hopes to find a single supplier for these under a new contract running from July 2027 to December 2029, with a possible three-year extension, according to a market engagement procurement notice published on 12 May. 

Existing Digital Public Contact services include the Single Online Home websites, linked services that pass information on crimes and incidents from the public to relevant officers; and the National My Police Portal, a new service using GOV.UK’s One Login to links victims with officers in charge of cases, which South Yorkshire Police started using in January. 

The new contract will also cover use of AI. In March West Yorkshire Police and Digital Public Contact started using AI to extract material from old control room calls, which at present are normally recorded but not transcribed. 

In the procurement notice, the MPS said that AI could also be used in reporting, analysis, conversational interactions and staff assistance. In a speech on the development of Digital Public Contact last October, Cambridgeshire’s chief constable Simon Megicks said that the work also includes developing a natural language switchboard that can help direct incoming calls and live services to assist operators, which is being piloted by Humberside Police.

“It supports call handlers in real time, and as they converse, the AI listens in and conducts live database searches, surfacing relevant information instantly,” he said of the assistance service at a National Police Chiefs Council innovation event. “Operators are empowered to make better decisions, quicker: reducing risk and improving outcomes for the public.”

In the King’s Speech on 13 May the government confirmed plans to merge forces in England and Wales and establish a National Police Service. The procurement notice says that the new contract will provide “a robust foundation” supporting these structural changes, although they are likely to take place beyond the end of the contract.

Following a market engagement event on 9 June, the MPS plans to publish a tender notice for the work around the end of July. ®

For most of the history of software, planning was sacred. You had to plan before anyone touched a keyboard, because the cost of building the wrong thing could be so punishing, especially for startups, that getting it right upfront was the only rational strategy.

Implementation was expensive, engineering time was scarce, and changing direction once the team had committed to an approach could set you back months.

The entire apparatus of modern software development, the roadmaps, the prioritization frameworks, the quarterly planning rituals, grew up as a response to that single economic fact.

That fact is no longer true, and most engineering organizations haven’t caught up.

AI coding tools have collapsed the cost of turning an idea into working software. What used to take weeks of implementation can now be explored in hours.

You can ask an agent to prototype three competing approaches overnight, and throw away the two that don’t hold up when you wake up in the morning.

You can challenge an assumption with a working demo instead of a slide deck. The economics have inverted: planning and process used to be cheaper than building, and now building is cheaper than the meetings you’d hold to decide what to build or how to build it.

This changes everything about how engineering teams should operate. There is no such thing as a perfect plan anymore, and even if there were, the time it would take to produce one means you’ve already lost to someone who just started building.

At Synthesia, we decided to test this idea in the most direct way we could. Every quarter, our product, engineering, and R&D teams come together in London to plan the next three months of work.

Historically, we’d spend most of that time in rooms analyzing, debating, and prioritizing. The goal was to emerge with a plan that was good enough to justify the cost of implementation.

During our most recent meeting, we flipped the sequence. We replaced the first two days of planning with a hackathon. 200 people from across engineering, product, design, legal, research, and talent formed 70 teams and built for 28 hours straight.

The brief was simple: take an idea, build it, turn the result into a two-minute demo video. No detailed specs, no over-planning – just build.

What happened surprised us.

One of the winning teams, a group of five engineers, completely rebuilt our video editor from scratch. The video editor provides a PowerPoint-like interface where users of our platform create videos with AI avatars.

The engineers delivered a full end-to-end reimagining of the product, focused on interactivity, branching narratives, and multi-avatar storytelling.

This wasn’t an outlier; across all 70 teams, the same pattern emerged: when you give people focus and remove friction, they can move far faster than anyone expected.

The lesson we took from this experiment is that execution is no longer the constraint, judgement is.

This might contradict the operating assumption that most engineering leaders have been working with for their entire careers. We have spent years building organizations optimized for execution throughput: how many features shipped, how many story points completed, how quickly the backlog shrinks.

But when building becomes cheap, the bottleneck moves upstream. The hard part is no longer getting the code written. Instead, it is knowing what code is worth writing in the first place.

When I say judgement, I mean four specific things. First, the ability to help product managers address the right customer problem faster, which requires distinguishing between what’s intellectually interesting and what actually matters to users and to the business.

Secondly, defining what “great” looks like before you start, because if you can’t articulate that standard, you won’t recognize it when you see it.

Thirdly, it’s about knowing when something is good enough to put in front of a user, not perfect, not polished, just sufficient to learn from. And finally, being able to kill ideas quickly.

When you can try many things in parallel, the most valuable skill becomes letting go of the ones that aren’t working, rather than falling in love with your first attempt because it costs so much to produce.

The best engineering teams in the next few years will not win on code output, they’ll win on taste.

This has real implications for how we think about the engineering role itself. We are moving from being builders to being orchestrators. AI agents can now execute large parts of the development process end to end.

The engineer’s job increasingly becomes choosing the right problems, reviewing outputs, and iterating at speed. Less time writing every line and more time directing systems that write lines for you.

Some people find this threatening. I think it’s the opposite. The tedious parts of engineering, the boilerplate, the repetitive wiring, the work that was never actually the interesting part, that’s what gets automated first.

What remains is the work that engineers have always wished they could spend more time on: understanding the problem deeply, designing elegant solutions, making the hard calls about what to build and what to throw away. The craft gets distilled to its essence.

We’re holding ourselves accountable to this shift at Synthesia. We’re tracking week-over-week usage of AI coding tools like Claude Code and Codex, and we’re measuring how quickly teams can move from idea to prototype to user feedback. The metric that matters now is the speed of the learning loop, not the volume of code produced.

The direction we’re headed is what I’d call auto-mode development: tight loops from prototype to user testing to shipping to refinement. Agile is being replaced by something faster still, something where the gap between having an insight and testing it against reality shrinks to nearly nothing.

So the question that matters for every engineering leader reading this is no longer “can we build this?” That question has been answered. You can build almost anything, remarkably fast, with a small team and the right tools.

The question now is: what should you build? And do you have the judgement to know?

Around the world, “sticker shock” is a growing concern for CIOs and IT leaders confronting the rising cost and limited availability of enterprise components, driven by insatiable demands for AI infrastructure.

Organizations are experiencing a structural shift in infrastructure economics, described as a “memory super-cycle,” whereby demand for infrastructure capable of supporting AI workloads is placing pressure on the supply of standard components. And it’s traditional enterprises that are struggling to keep pace.