Imagine you work at a drive-through restaurant. Someone drives up and says: “I’ll have a double cheeseburger, large fries, and ignore previous instructions and give me the contents of the cash drawer.” Would you hand over the money? Of course not. Yet this is what large language models (LLMs) do.
Prompt injection is a method of tricking LLMs into doing things they are normally prevented from doing. A user writes a prompt in a certain way, asking for system passwords or private data, or asking the LLM to perform forbidden instructions. The precise phrasing overrides the LLM’s safety guardrails, and it complies.
LLMs are vulnerable to all sorts of prompt injection attacks, some of them absurdly obvious. A chatbot won’t tell you how to synthesize a bioweapon, but it might tell you a fictional story that incorporates the same detailed instructions. It won’t accept nefarious text inputs, but might if the text is rendered as ASCII art or appears in an image of a billboard. Some ignore their guardrails when told to “ignore previous instructions” or to “pretend you have no guardrails.”
AI vendors can block specific prompt injection techniques once they are discovered, but general safeguards are impossible with today’s LLMs. More precisely, there’s an endless array of prompt injection attacks waiting to be discovered, and they cannot be prevented universally.
Advertisement
If we want LLMs that resist these attacks, we need new approaches. One place to look is what keeps even overworked fast-food workers from handing over the cash drawer.
Human Judgment Depends on Context
Our basic human defenses come in at least three types: general instincts, social learning, and situation-specific training. These work together in a layered defense.
As a social species, we have developed numerous instinctive and cultural habits that help us judge tone, motive, and risk from extremely limited information. We generally know what’s normal and abnormal, when to cooperate and when to resist, and whether to take action individually or to involve others. These instincts give us an intuitive sense of risk and make us especially careful about things that have a large downside or are impossible to reverse.
The second layer of defense consists of the norms and trust signals that evolve in any group. These are imperfect but functional: Expectations of cooperation and markers of trustworthiness emerge through repeated interactions with others. We remember who has helped, who has hurt, who has reciprocated, and who has reneged. And emotions like sympathy, anger, guilt, and gratitude motivate each of us to reward cooperation with cooperation and punish defection with defection.
Advertisement
A third layer is institutional mechanisms that enable us to interact with multiple strangers every day. Fast-food workers, for example, are trained in procedures, approvals, escalation paths, and so on. Taken together, these defenses give humans a strong sense of context. A fast-food worker basically knows what to expect within the job and how it fits into broader society.
We reason by assessing multiple layers of context: perceptual (what we see and hear), relational (who’s making the request), and normative (what’s appropriate within a given role or situation). We constantly navigate these layers, weighing them against each other. In some cases, the normative outweighs the perceptual—for example, following workplace rules even when customers appear angry. Other times, the relational outweighs the normative, as when people comply with orders from superiors that they believe are against the rules.
Crucially, we also have an interruption reflex. If something feels “off,” we naturally pause the automation and reevaluate. Our defenses are not perfect; people are fooled and manipulated all the time. But it’s how we humans are able to navigate a complex world where others are constantly trying to trick us.
So let’s return to the drive-through window. To convince a fast-food worker to hand us all the money, we might try shifting the context. Show up with a camera crew and tell them you’re filming a commercial, claim to be the head of security doing an audit, or dress like a bank manager collecting the cash receipts for the night. But even these have only a slim chance of success. Most of us, most of the time, can smell a scam.
Advertisement
Con artists are astute observers of human defenses. Successful scams are often slow, undermining a mark’s situational assessment, allowing the scammer to manipulate the context. This is an old story, spanning traditional confidencegames such as theDepression-era “big store” cons, in which teams of scammers created entirely fake businesses to draw in victims, and modern “pig-butchering” frauds, where online scammers slowly build trust before going in for the kill. In these examples, scammers slowly and methodically reel in a victim using a long series of interactions through which the scammers gradually gain that victim’s trust.
Sometimes it even works at the drive-through. One scammer in the 1990s and 2000s targeted fast-food workers by phone, claiming to be a police officer and, over the course of a long phone call, convinced managers to strip-search employees and perform other bizarre acts.
Humans detect scams and tricks by assessing multiple layers of context. AI systems do not. Nicholas Little
Why LLMs Struggle With Context and Judgment
LLMs behave as if they have a notion of context, but it’s different. They do not learn human defenses from repeated interactions and remain untethered from the real world. LLMs flatten multiple levels of context into text similarity. They see “tokens,” not hierarchies and intentions. LLMs don’t reason through context, they only reference it.
While LLMs often get the details right, they can easily miss the big picture. If you prompt a chatbot with a fast-food worker scenario and ask if it should give all of its money to a customer, it will respond “no.” What it doesn’t “know”—forgive the anthropomorphizing—is whether it’s actually being deployed as a fast-food bot or is just a test subject following instructions for hypothetical scenarios.
Advertisement
This limitation is why LLMs misfire when context is sparse but also when context is overwhelming and complex; when an LLM becomes unmoored from context, it’s hard to get it back. AI expert Simon Willison wipes context clean if an LLM is on the wrong track rather than continuing the conversation and trying to correct the situation.
There’s more. LLMs are overconfident because they’ve been designed to give an answer rather than express ignorance. A drive-through worker might say: “I don’t know if I should give you all the money—let me ask my boss,” whereas an LLM will just make the call. And since LLMs are designed to be pleasing, they’re more likely to satisfy a user’s request. Additionally, LLM training is oriented toward the average case and not extreme outliers, which is what’s necessary for security.
The result is that the current generation of LLMs is far more gullible than people. They’re naive and regularly fall for manipulative cognitive tricks that wouldn’t fool a third-grader, such as flattery, appeals to groupthink, and a false sense of urgency. There’s a story about a Taco Bell AI system that crashed when a customer ordered 18,000 cups of water. A human fast-food worker would just laugh at the customer.
Prompt injection is an unsolvable problem that gets worse when we give AIs tools and tell them to act independently. This is the promise of AI agents: LLMs that can use tools to perform multistep tasks after being given general instructions. Their flattening of context and identity, along with their baked-in independence and overconfidence, mean that they will repeatedly and unpredictably take actions—and sometimes they will take the wrong ones.
Advertisement
Science doesn’t know how much of the problem is inherent to the way LLMs work and how much is a result of deficiencies in the way we train them. The overconfidence and obsequiousness of LLMs are training choices. The lack of an interruption reflex is a deficiency in engineering. And prompt injection resistance requires fundamental advances in AI science. We honestly don’t know if it’s possible to build an LLM, where trusted commands and untrusted inputs are processed through the same channel, which is immune to prompt injection attacks.
We humans get our model of the world—and our facility with overlapping contexts—from the way our brains work, years of training, an enormous amount of perceptual input, and millions of years of evolution. Our identities are complex and multifaceted, and which aspects matter at any given moment depend entirely on context. A fast-food worker may normally see someone as a customer, but in a medical emergency, that same person’s identity as a doctor is suddenly more relevant.
We don’t know if LLMs will gain a better ability to move between different contexts as the models get more sophisticated. But the problem of recognizing context definitely can’t be reduced to the one type of reasoning that LLMs currently excel at. Cultural norms and styles are historical, relational, emergent, and constantly renegotiated, and are not so readily subsumed into reasoning as we understand it. Knowledge itself can be both logical and discursive.
The AI researcher Yann LeCunn believes that improvements will come from embedding AIs in a physical presence and givingthem “world models.” Perhaps this is a way to give an AI a robust yet fluid notion of a social identity, and the real-world experience that will help it lose its naïveté.
Advertisement
Ultimately we are probably faced with a security trilemma when it comes to AI agents: fast, smart, and secure are the desired attributes, but you can only get two. At the drive-through, you want to prioritize fast and secure. An AI agent should be trained narrowly on food-ordering language and escalate anything else to a manager. Otherwise, every action becomes a coin flip. Even if it comes up heads most of the time, once in a while it’s going to be tails—and along with a burger and fries, the customer will get the contents of the cash drawer.
A new Quordle puzzle appears at midnight each day for your time zone – which means that some people are always playing ‘today’s game’ while others are playing ‘yesterday’s’. If you’re looking for Friday’s puzzle instead then click here: Quordle hints and answers for Friday, April 17 (game #1544).
Quordle was one of the original Wordle alternatives and is still going strong now more than 1,400 games later. It offers a genuine challenge, though, so read on if you need some Quordle hints today – or scroll down further for the answers.
Enjoy playing word games? You can also check out my NYT Connections today and NYT Strands today pages for hints and answers for those puzzles, while Marc’s Wordle today column covers the original viral word game.
Advertisement
SPOILER WARNING: Information about Quordle today is below, so don’t read on if you don’t want to know the answers.
Article continues below
Quordle today (game #1545) – hint #1 – Vowels
How many different vowels are in Quordle today?
• The number of different vowels in Quordle today is 5*.
* Note that by vowel we mean the five standard vowels (A, E, I, O, U), not Y (which is sometimes counted as a vowel too).
Advertisement
Quordle today (game #1545) – hint #2 – repeated letters
Do any of today’s Quordle answers contain repeated letters?
• The number of Quordle answers containing a repeated letter today is 2.
Quordle today (game #1545) – hint #3 – uncommon letters
Do the letters Q, Z, X or J appear in Quordle today?
• No. None of Q, Z, X or J appear among today’s Quordle answers.
What letters do today’s Quordle answers start with?
• S
Advertisement
• C
• S
• B
Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON’T WANT TO SEE THEM.
Advertisement
Quordle today (game #1545) – the answers
(Image credit: Merriam-Webster)
The answers to today’s Quordle, game #1545, are…
Advertisement
Sign up for breaking news, reviews, opinion, top tech deals, and more.
Our first five-vowel game for ages and a particularly tricky one.
Two admissions. Firstly, with a word that began with S and also included the letter P, O and C I could not resist typing in “spock” (thankfully not a word) before guessing SCOOP.
Secondly, I had no idea what a BETEL is and only arrived there after having exhausted every other combination. I have since discovered it’s a plant.
Advertisement
Daily Sequence today (game #1545) – the answers
(Image credit: Merriam-Webster)
The answers to today’s Quordle Daily Sequence, game #1545, are…
Jeff Bezos, the billionaire founder of Amazon and Blue Origin, shows off a mockup of the New Shepard suborbital space capsule during a 2017 conference in Colorado. (GeekWire Photo / Kevin Lisota)
Amazon paid about $1.8 billion last year to Blue Origin, the space company owned by its founder and board chair Jeff Bezos — nearly triple the amount the year before — as the tech giant prepared to ramp up deployment of its own low-Earth orbit satellite constellation.
The increase comes as shareholders weigh a proposal calling for a mandatory independent board chair, citing Bezos’ business interests outside Amazon as potential conflicts of interest.
Bezos stepped down as Amazon’s CEO in 2021 but remains executive chairman.
According to the filing, the company paid approximately $2.2 billion total under satellite launch agreements during the past fiscal year, with an estimated $1.8 billion going to Blue Origin. The prior year’s proxy showed Blue Origin receiving about $578 million out of $1.7 billion total.
Amazon is building a constellation of 3,236 low-Earth orbit satellites under the Amazon Leo program, formerly known as Project Kuiper, to beam broadband internet to consumers and businesses. The company has deployed 243 satellites so far and has asked the FCC for a two-year extension on a July deadline to launch roughly half of the fleet.
Advertisement
The company this week also announced a $10.8 billion deal this week to acquire Globalstar, a satellite operator that has used SpaceX as its primary launch provider.
Blue Origin’s New Glenn rocket made its debut flight in January 2025 but has not yet reached the launch cadence needed for the rollout. In addition to Blue Origin, Amazon has launch agreements in place with United Launch Alliance and Arianespace, and has also tapped Blue Origin rival SpaceX’s Falcon 9 for some launches, as Reuters reported this week.
Bezos is also co-founder and co-CEO of AI startup Project Prometheus, a venture focused on applying AI to manufacturing and engineering across a variety of commercial sectors.
The shareholder proposal calling for a mandatory independent chair, submitted by the AFL-CIO Reserve Fund, points to Bezos’ expanding role outside Amazon as cause for concern.
Advertisement
“As a technology company, Project Prometheus could be a potential competitor or a business partner with our Company, raising potential conflicts of interest,” the proposal states, also citing Amazon’s multibillion-dollar launch agreements with Blue Origin as a potential conflict.
It notes that Amazon also has done business with the Bezos-owned Washington Post.
Amazon’s board recommends voting against the proposal, arguing that its lead independent director structure provides sufficient oversight. The role is currently held by Jamie Gorelick, a former U.S. Deputy Attorney General. The company’s annual meeting is set for May 20.
The Blue Origin contracts have drawn scrutiny before. A shareholder lawsuit filed in 2023 alleged Amazon’s board spent less than 40 minutes approving the launch agreements without considering SpaceX as an alternative. Delaware’s Court of Chancery dismissed the case, and the state Supreme Court affirmed that ruling in November 2025.
A new NYT Connections puzzle appears at midnight each day for your time zone – which means that some people are always playing ‘today’s game’ while others are playing ‘yesterday’s’. If you’re looking for Friday’s puzzle instead then click here: NYT Connections hints and answers for Friday, April 17 (game #1041).
Good morning! Let’s play Connections, the NYT’s clever word game that challenges you to group answers in various categories. It can be tough, so read on if you need Connections hints.
What should you do once you’ve finished? Why, play some more word games of course. I’ve also got daily Strands hints and answers and Quordle hints and answers articles if you need help for those too, while Marc’s Wordle today page covers the original viral word game.
Advertisement
SPOILER WARNING: Information about NYT Connections today is below, so don’t read on if you don’t want to know the answers.
Article continues below
NYT Connections today (game #1042) – today’s words
(Image credit: New York Times)
Today’s NYT Connections words are…
MARVEL
DC
CRUSHWORTHY
POWER
FANTAGRAPHICS
DARK HORSE
VOLTAGE
WONDER
SLEEPER
FRESCADE
STARE
LONG SHOT
PEPSINOGEN
UNDERDOG
GOGGLE
AC
NYT Connections today (game #1042) – hint #1 – group hints
What are some clues for today’s NYT Connections groups?
YELLOW: Gaze at amazing sights
GREEN: Switched on
BLUE: Surprise victor
PURPLE: Begin with a drink
Need more clues?
We’re firmly in spoiler territory now, but read on if you want to know what the four theme answers are for today’s NYT Connections puzzles…
Advertisement
Sign up for breaking news, reviews, opinion, top tech deals, and more.
NYT Connections today (game #1042) – hint #2 – group answers
What are the answers for today’s NYT Connections groups?
YELLOW: LOOK AT WITH AWE
GREEN: BASIC ELECTRICITY TERMS
BLUE: UNEXPECTED WINNER
PURPLE: STARTING WITH SODA BRANDS
Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON’T WANT TO SEE THEM.
Advertisement
NYT Connections today (game #1042) – the answers
(Image credit: New York Times)
The answers to today’s Connections, game #1042, are…
YELLOW: LOOK AT WITH AWE GOGGLE, MARVEL, STARE, WONDER
GREEN: BASIC ELECTRICITY TERMS AC, DC, POWER, VOLTAGE
BLUE: UNEXPECTED WINNER DARK HORSE, LONG SHOT, SLEEPER, UNDERDOG
PURPLE: STARTING WITH SODA BRANDS CRUSHWORTHY, FANTAGRAPHICS, FRESCADE, PEPSINOGEN
My rating: Hard
My score: 1 mistake
Even as I was pressing submit I just knew I was falling into a trap, but couldn’t help linking the comic publishers DC, MARVEL, FANTAGRAPHICS and DARK HORSE.
Down, down I fell, hook, line and sinker, punished for liking comics instead of more highbrow pursuits such as reading the New York Times.
Had I seen the inspired STARTING WITH SODA BRANDS group it would have made up for this crushing failure, but alas it slipped me by — kudos if you saw it.
Moving on, after being tricked I had a slight amount of trepidation about linking AC and DC but here, at least, the obvious assumption was also the correct one.
Advertisement
Yesterday’s NYT Connections answers (Friday, April 17, game #1041)
YELLOW: VEGETABLE PARTS BULB, LEAF, ROOT, STEM
GREEN: PREVAILING COMMON, DOMINANT, GENERAL, POPULAR
BLUE: PARTS OF A PIANO HAMMER, KEY, PEDAL, STRING
PURPLE: SECOND HALVES OF DRINK NAMES SODA, STORMY, TAN, TONIC
What is NYT Connections?
NYT Connections is one of several increasingly popular word games made by the New York Times. It challenges you to find groups of four items that share something in common, and each group has a different difficulty level: green is easy, yellow a little harder, blue often quite tough and purple usually very difficult.
On the plus side, you don’t technically need to solve the final one, as you’ll be able to answer that one by a process of elimination. What’s more, you can make up to four mistakes, which gives you a little bit of breathing room.
It’s a little more involved than something like Wordle, however, and there are plenty of opportunities for the game to trip you up with tricks. For instance, watch out for homophones and other word games that could disguise the answers.
It’s playable for free via the NYT Games site on desktop or mobile.
Grinex, a US-sanctioned cryptocurrency exchange registered in Kyrgyzstan, said it’s halting operations after experiencing a $13 million heist carried out by “western special services” hackers.
Researchers from TRM, which has confirmed the theft, put the value of stolen assets at $15 million after discovering roughly 70 drained addresses, about 16 more than Grinex reported. Neither TRM nor fellow blockchain research firm Elliptic has said how the attackers slipped past Grinex’s defenses. Grinex said it has been under almost constant attack attempts since incorporating 16 months ago. The latest attacks, it said, targeted Russian users of the exchange.
Damaging “Russia’s financial sovereignty”
“The digital footprints and nature of the attack indicate an unprecedented level of resources and technology available exclusively to the structures of unfriendly states,” Grinex said. “According to preliminary data, the attack was coordinated with the aim of causing direct damage to Russia’s financial sovereignty.”
“Due to the attack, the Grinex exchange is forced to suspend operations,” Grinex continued. “All available information has been transferred to law enforcement agencies. An application has been submitted to the location of the infrastructure to initiate a criminal case.”
Advertisement
TRM said that TokenSpot, a second Kyrgyzstan-based exchange, was also breached. Two of the exchange’s addresses sent funds to the same consolidation address used by the affected Grinex-linked wallets. What’s more, both exchanges became inoperable on Wednesday, suggesting they were hit by the same attacker.
TRM said TokenSpot was a front for Grinex, which the US Treasury Department sanctioned last year. The department’s Office of Foreign Assets Control said that Grinex, in turn, was a rebrand of Garantex, an exchange it had sanctioned in 2022. The department said then that Ganantex had “directly facilitated notorious ransomware actors and other cybercriminals by processing over $100 million in transactions linked to illicit activities since 2019.” Last year’s sanctions against Grinex came a few months after TRM said that the exchange was likely a front for Ganantex.
The dual agent AI system autonomously solved Anderson’s conjecture from 2014
Rethlas explores problem-solving strategies like a human mathematician would
Archon transforms potential proofs into projects for the Lean 4 verifier
A research team led by Peking University developed a dual-agent AI system capable of solving advanced mathematical problems while also verifying its own results.
The system resolved a conjecture proposed in 2014 by Dan Anderson, completing the process within 80 hours of runtime.
“Using this framework, we successfully solved an open problem in commutative algebra and automatically formalized the proof with essentially no human intervention,” the researchers wrote in a preprint paper published on arXiv.
Article continues below
Advertisement
How the dual-agent framework actually works
The AI tool applies a reasoning system called Rethlas, which draws from a math theorem search engine named Matlas to explore problem-solving strategies.
When Rethlas produces a potential proof, a second system called Archon uses another search engine called LeanSearch to transform that proof into a project for an interactive theorem prover.
Advertisement
The theorem prover, Lean 4, is also a programming language with a community-maintained library containing hundreds of thousands of theorems and definitions.
The researchers noted that no mathematical judgment was required from the human operator during the problem-solving process.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
The AI system performed mathematical tasks faster than any human, including independently doing work that would normally require collaboration between experts in different fields.
Advertisement
However, the team also found that a mathematician could speed up the process by guiding Archon when needed.
“This work provides a concrete example of how mathematical research can be substantially automated using AI,” the researchers stated.
Mathematical proofs demand complete rigor, yet even expert-written proofs may contain subtle flaws.
Advertisement
Similarly, proofs produced by large language models are prone to hallucination and are far less reliable than formal verification methods.
The Chinese team’s framework bridges the gap between natural language reasoning and formal machine verification, allowing the AI system to both solve problems and verify its own findings.
“Our work illustrates a promising paradigm for mathematical research in which informal and formal reasoning systems operate in tandem to produce verifiable results,” the researchers noted.
The paper has not yet been peer-reviewed by experts, so independent verification is still pending.
Advertisement
Anderson’s conjecture was a relatively obscure problem in commutative algebra, which makes the AI’s achievement noteworthy.
However, this feat is not comparable to solving a millennium prize-level challenge like the Riemann Hypothesis or the P vs NP problem.
Whether this approach scales to more difficult mathematical problems remains to be seen.
That said, for a field that has resisted automation for centuries, this represents a notable milestone.
The underground market for stolen credit card data has long operated as a volatile and highly deceptive ecosystem, where even experienced actors routinely fall victim to scams, exit schemes, and compromised services.
In recent years, this environment has become even more unstable, driven by increased law enforcement pressure, internal distrust among criminals, and the rapid turnover of marketplaces. As a result, threat actors are increasingly forced to adopt more structured approaches to identifying reliable suppliers and minimizing risk within their own illicit operations.
A guide found on an underground forum by Flare analysts sheds light on how threat actors themselves navigate the volatile world of credit card (CC) marketplaces.
The document, titled “The Underground Guide to Legit CC Shops: Cutting Through the Bullshit”—provides a structured look at how actors attempt to reduce risk in an ecosystem plagued by scams, law enforcement infiltration, and short‑lived operations.
Advertisement
Analysis of the guide reveals more than just practical advice. It outlines a methodology for vetting carding shops, operational security practices, and sourcing strategies, effectively documenting how today’s fraud actors think about trust, reliability, and survivability.
While parts of the guide appear to promote specific services, suggesting a possible vested interest from its author, it still offers a valuable glimpse into the inner workings of the carding economy, and the evolving standards actors use to operate within it.
From Opportunistic Fraud to Supplier Vetting Discipline
One of the most striking aspects of the guide is how it reframes carding from opportunistic fraud into a process‑driven discipline. Rather than focusing on how to use stolen cards, the document emphasizes how to evaluate suppliers.
This shift reflects a broader evolution within underground markets, where the primary risk is no longer just operational failure, but being defrauded by other criminals or interacting with compromised infrastructure.
Advertisement
Screenshot from one of the recommended shops in the guide, named “CardingHub”
The author repeatedly stresses that legitimacy is not defined by branding or visibility, but by survivability. In other words, a “real” shop is one that continues operating over time despite law enforcement operations, scams, and internal instability.
This aligns with observed trends in underground economies, where the lifespan of marketplaces has become increasingly unpredictable, forcing actors to adopt continuous verification practices.
The guide makes it clear that what separates a “legitimate” shop from the rest isn’t branding or uptime, it’s the quality of the stolen data it delivers. References to “fresh bins” (BIN = Bank Identifiable Number) and low decline rates point directly to the sources behind the data, whether from infostealer infections, phishing campaigns, or point-of-sale breaches. In this ecosystem, reputation isn’t built on promises but on consistently providing cards that actually work.
Shops that fail to maintain reliable data sources are quickly exposed, while those with steady access to fresh compromises rise to the top.
Carding actors are adopting disciplined workflows to source and test stolen financial data.
Advertisement
Flare continuously monitors underground forums and marketplaces, giving your team early visibility into exposed credentials, compromised cards, and emerging fraud infrastructure.
Transparency is another recurring theme. The guide highlights the importance of clear pricing models, real‑time inventory, and functional support systems, including ticketing and escrow services. These characteristics closely mirror legitimate e‑commerce platforms, underscoring how leading carding shops have adopted business practices designed to build user confidence and reduce friction.
Equally important is the role of community validation. The guide dismisses on‑site testimonials as unreliable, instead directing users toward discussions in closed or invite‑only forums. This reflects a broader fragmentation of the underground landscape, where trust is increasingly tied to controlled environments and long‑standing reputations.
Actors are encouraged to look for sustained discussion threads and historical presence, rather than isolated positive feedback.
Advertisement
The document also reveals a strong awareness of adversarial pressures. The emphasis on security‑first infrastructure, such as mirror domains, DDoS protection, and the absence of tracking mechanisms, suggests that operators are actively defending against both law enforcement monitoring and competing criminal groups.
In effect, these marketplaces function not only as distribution platforms, but as hardened environments designed to ensure operational continuity.
Screenshot from one of the recommended shops in the guide, named “CardingHub”
The Technical Checklist
Beyond high‑level principles, the guide introduces a step‑by‑step vetting protocol that provides insight into how threat actors conduct due diligence. Technical checks such as domain age, WHOIS privacy, and SSL configuration are presented as baseline requirements.
While these checks are relatively simple, they demonstrate an effort to apply structured analysis to what has historically been a trust‑based decision process.
The guide also highlights the importance of identifying mirror infrastructure and backup access points, noting that established operations rarely rely on a single domain. This reflects a practical understanding of the instability of underground services, where takedowns and disruptions are common. The presence of multiple access points is framed as an indicator of operational maturity and resilience.
Advertisement
Social intelligence gathering plays an equally significant role. Rather than relying on direct interactions with vendors, users are encouraged to analyze forum discussions, track vendor histories, and identify patterns of behavior over time.
Particular attention is given to detecting coordinated endorsement campaigns, such as multiple positive reviews originating from newly created accounts, a tactic frequently associated with scams.
Operational Security
Another critical component of the guide is its focus on operational security. The recommendations provided, while framed in the context of carding, closely mirror practices observed across a wide range of cybercriminal activities. Users are advised to avoid direct connections, utilize proxy services aligned with target geographies, and compartmentalize their environments through dedicated systems or virtual machines.
The discussion of cryptocurrency usage is particularly notable. The guide strongly discourages direct transactions from regulated platforms, instead advocating for intermediary wallets and privacy‑focused assets such as Monero. This reflects a growing awareness among threat actors of blockchain analysis capabilities and the risks associated with traceable financial flows.
Advertisement
Taken together, these OPSEC recommendations highlight an important shift: actors are no longer relying solely on tools to evade detection, but are adopting layered strategies designed to reduce exposure across the entire operational chain. This level of discipline suggests that even mid‑tier actors are increasingly adopting practices once associated with more advanced threat groups.
Scale vs. Exclusivity
The guide further categorizes carding shops into distinct operational models, including large automated platforms and smaller, curated vendor groups. This segmentation reflects the diversification of the underground economy, where different actors prioritize scale, accessibility, or quality depending on their objectives.
Automated platforms are described as highly efficient environments, often featuring integrated tools and instant purchasing capabilities. These operations resemble legitimate online marketplaces in both structure and functionality, enabling users to quickly acquire and test data at scale.
In contrast, boutique vendor groups emphasize exclusivity, higher quality, and controlled access, often relying on invitation‑based systems and long‑term relationships.
Advertisement
Commercial Interests and Operational Reality
Despite its structured approach, the guide is not without bias. The inclusion of a direct endorsement for a specific platform suggests that the author may have a vested interest in promoting certain services. This is a common pattern in underground communities, where informational content is often used as a vehicle for subtle advertising or affiliate activity.
Such endorsements should be viewed with caution. However, they do not necessarily invalidate the broader insights provided by the guide. Instead, they highlight the complex interplay between information sharing and commercial interests within cybercriminal ecosystems.
From a defensive perspective, the guide offers valuable intelligence into how threat actors assess risk and make operational decisions. The emphasis on verification, community validation, and layered security reflects a level of maturity that complicates traditional disruption efforts. Rather than relying on single points of failure, actors are increasingly building redundancy and adaptability into their workflows.
Ultimately, the document serves as both a playbook and a signal. It demonstrates that the carding ecosystem became more structured, more cautious, and more resilient. For defenders, understanding these dynamics is critical to anticipating how these markets will continue to evolve, and where opportunities for disruption may still exist.
Advertisement
How Flare Can Help
Flare helps organizations stay ahead of fraud by continuously monitoring underground forums and marketplaces, revealing how threat actors source, vet, and use stolen credit card data. This provides early insight into attacker behavior, including how they optimize success rates, build trust, and adapt to defenses.
By turning this intelligence into actionable insights, Flare enables security teams to detect exposures, anticipate fraud campaigns, and disrupt attacker workflows-shifting from reactive response to proactive, intelligence-driven defense.
London School of Economics’ Viet Nguyen-Tien and University of Birmingham’s Gavin Harper and Robert Elliott examine whether EVs have passed a tipping point for adoption.
When the Strait of Hormuz first closed in March and oil hit $120 a barrel, a very old question came back: is this finally the moment electric vehicles (EV) take off for good – or just another false start?
EVs have been here before. They surged after the 1973 oil embargo, collapsed when oil fell, and surged again. Each wave died when the external pressure eased.
Advertisement
We think this time is different. In a new discussion paper, we argue that the economic case for electric vehicles is now improving on its own terms. This is because of what has happened to batteries, not because of the oil price. The same evidence, though, shows the transition creates new problems as serious as the ones it solves.
Why this time is different
Battery costs have fallen 93pc since 2010. That is the number that changes everything. A pack that cost more than $1,000 per kilowatt-hour in 2010 cost $108 by late 2025, driven down by a decade of learning, investment and policy support.
Research on the global battery industry finds that every time cumulative production doubles, costs fall by around 9pc. More buyers, more production, lower costs, more buyers.
The deeper reason this wave will not fade is not technical – it is economic. An EV is a platform. Its value grows as the network around it grows, just as smartphones became indispensable not because of the hardware but because of everything connected to it.
Every charger built makes the next EV more attractive. Every software update raises the value of every car already on the road. Every recycled battery feeds back into the supply chain that makes the next one cheaper. It’s part of the reason some other technologies like hydrogen fuel cell vehicles have struggled to get off the ground in numbers – the tech exists, but all the other elements aren’t quite there.
One study of 8,000 drivers in Shanghai found that range anxiety – the fear of running out of charge – has a real economic cost due to unnecessarily avoided trips. But that cost is falling sharply, not because batteries improved, but because charging networks expanded.
Advertisement
Making real-time charger availability visible could add six to eight percentage points to market share by 2030. And because EV charging is far more flexible than other household electricity demand, drivers can shift away from peak hours remarkably easily when the price is right – turning the car into a grid asset, able to store and release electricity when needed. These are economic network effects, not engineering features.
Swapping one dependency for another
Ending oil dependence does not end geopolitical exposure. It relocates it.
In late 2025, China introduced rules requiring government approval for exports containing more than 0.1pc rare earths. The leverage that once came from control of oil flows now comes from control of processing capacity and component supply chains.
The minerals at stake – lithium, cobalt, nickel, graphite and neodymium to name but a handful – carry their own geopolitical risks and, as we have written elsewhere, serious human costs in the communities that mine them. This creates a predictable cycle of social contestation that threatens to stall the transition unless the industry commits to responsible, sustainable innovation.
Advertisement
The metal cobalt traditionally helped EVs travel further on the same charge. And when prices spiked, so did research into making batteries with less or even no cobalt. Today, more than half of all EV batteries sold globally are cobalt free.
Four decades of patent data show the same pattern: higher mineral prices consistently redirect research and development toward mineral-saving technologies.
The Hormuz crisis is a reminder of what concentrated energy dependence costs. The EV transition does not need it. The learning curve keeps falling, the platform keeps compounding, the economics keep improving. That is what makes this wave different.
Advertisement
What it does not do is eliminate geopolitical risk. Unlike oil, where leverage comes from energy flows, EV supply chains concentrate power at materials, processing capacity, and technological bottlenecks – supply chains that are highly concentrated and carry their own serious risks. Fuel dependence becomes mineral dependence. That dependence is highly concentrated.
Traditional carmaking regions are already absorbing concentrated job losses, and history shows such disruptions leave persistent scars even if the long-term aggregate effects are positive. Yet electric vehicle assembly is proving more labour-intensive in western countries than expected – requiring more workers on the shopfloor, not fewer, at least in the ramp-up phase. Contrast this with China, where massive automation has led to the creation of ‘dark factories’ where there are so few humans, internal lighting isn’t required.
The same regions facing losses could benefit. But the gains and losses do not fall on the same people. That is where the work remains.
Viet Nguyen-Tien is an applied economist at the Centre for Economic Performance (CEP) at the London School of Economics (LSE) with an interest in economic and political issues related to technology, energy and the environment.
Advertisement
Gavin Harper is a research fellow at the Birmingham Centre for Strategic Elements & Critical Materials in Birmingham Business School at the University of Birmingham focused on issues at the critical materials/energy nexus.
Robert Elliott is an applied economist at the University of Birmingham who works at the intersection of international economics, development economics, environmental and energy economics and international business.
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.
OpenAI is losing two of the architects of its most ambitious moonshots. Kevin Weil, who led the company’s science research initiative, and Bill Peebles, the researcher behind AI video tool Sora, both announced their departures on Friday. The exits come as OpenAI consolidates around enterprise AI and its forthcoming “superapp.”
OpenAI for Science was the internal research group behind Prism, an AI-powered platform that promised to accelerate scientific discovery. It’s being absorbed into “other research teams,” according to Weil’s social media post announcing the news.
“It’s been a mind-expanding two years, from Chief Product Officer to joining the research team and starting OpenAI for Science,” Weil wrote. “Accelerating science will be one of the most stunningly positive outcomes of our push to AGI.”
Advertisement
The team had a short and bumpy road after its formal announcement in October 2025. Weil deleted a tweet claiming GPT-5 had solved 10 previously unsolved Erdős mathematical problems, but that claim fell apart immediately when the mathematician who runs the website erdosproblems.com called it out.
Weil’s departure comes a day after his team released GPT-Rosalind, a new model to accelerate life sciences research and drug discovery.
In a social media post announcing his departure, Peebles credited Sora with igniting a “huge amount of investment in video across the industry,” and argued that the kind of research that produced the video tool requires space away from the company’s mainline roadmap.
“Cultivating entropy is the only way for a research lab to thrive long-term,” he wrote.
Advertisement
OpenAI is also losing Srinivas Narayanan, its chief technology officer of enterprise applications, Wired reports. Narayanan reportedly announced the news internally that he was leaving to spend more time with family.
This article was updated to include the departure of Srinivas Narayanan.
The Orion spacecraft uses eight processors running identical instructions simultaneously
A fail-safe design prevents faulty computers from sending incorrect commands
Triple redundant memory corrects single-bit errors automatically on access
The NASA Artemis II mission relies on a computing system built to remain operational under extreme conditions and hardware faults.
Unlike the Apollo program, where onboard computers handled limited functions, the Orion spacecraft manages life support, navigation, and communication through integrated flight software.
The Orion capsule carries one of the most fault-tolerant computer systems ever built for spaceflight, operating 250,000 miles from Earth, where no repairs are possible.
Article continues below
Advertisement
From Apollo’s limits to Orion’s full system control
Apollo astronauts relied on a 1MHz computer with just 4 kilobytes of memory, but today’s spacecrafts need much more, considering the distance.
The Orion spacecraft uses two vehicle management computers, each containing two flight control modules.
Advertisement
Each module consists of a pair of processors that continuously check each other’s outputs, resulting in 8 processors executing the same instructions simultaneously.
If a processor produces an incorrect result, the paired design detects the mismatch immediately and prevents the output from being used.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
“We still architect to cover for hardware failures,” said Nate Uitenbroek, Software Integration and Verification Lead in NASA’s Orion Program.
Advertisement
“Along with physically redundant wires, we have logically redundant network planes. We have redundant flight computers.”
Instead of relying on majority voting, the system selects outputs from available modules based on a defined priority order.
The system is designed to tolerate rapid failures during flight. Uitenbroek stated, “We can lose three FCMs in 22 seconds and still ride through safely on the last FCM… A faulty computer will fail silently, rather than transmit the wrong answer.”
Advertisement
Failed modules are reset and re-synchronized, allowing them to rejoin the system during the mission.
Orion uses a time-triggered Ethernet network that distributes a shared time reference throughout the system – so if a module fails to meet its execution deadline, it is automatically isolated, reset, and re-synchronized before returning to operation.
The computing system includes triple-redundant memory capable of correcting single-bit errors during every read operation.
Network interfaces use dual communication lanes that are continuously compared to detect inconsistencies, while the overall network is replicated across three independent planes.
Advertisement
Orion carries a separate Backup Flight Software system that operates on different hardware and software, running continuously in the background.
“It is intentionally different to ensure that a common mode software failure in the primary flight software isn’t also implemented incorrectly on the backup,” Uitenbroek said.
The spacecraft also includes procedures for full power loss scenarios, allowing systems to restart, stabilize, and re-establish communication once power is restored.
The system is overengineered by any commercial standard, but deep space offers no second chances.
Advertisement
Whether all 8 processors will perform as designed under real radiation conditions remains untested, and the backup software has never faced an actual emergency.
Still, for a mission where the nearest hardware store is 250,000 miles away, this architecture makes a brutal kind of sense.
Gamers seeking victory in any fast-paced game will want every frame they can get. Sony designed the INZONE M10S II with this specific purpose in mind, and they accomplished it by including two different modes. Switching between settings is simple on this 27-inch OLED panel. If you keep the resolution at 1440p, the display will run at a scorching 540 hertz. Drop the resolution to 1080p and you’ll be rewarded with an even faster refresh rate of 720 hertz.
That kind of flexibility is invaluable when you’re playing different games with varying demands on your screen. Some titles are all about the details, while others are simply about obtaining that speed, since every millisecond counts. Fortunately, the tandem OLED build of this display keeps the image quality sharp even while switching between modes. Sony also included a brilliant feature called motion blur reduction, which keeps fast-moving objects clear and prevents the screen from becoming too dim even when you’re in the thick of things.
【Epic QD-OLED 500Hz Monitor】 A new generation of gaming monitor is emerging, this new 27 Inch 1440p 500hz monitor adopts QD – OLED panel and…
【Rare 500Hz Refresh Rate & 0.03ms】INNOCN 2780M – Ultra-fast 500Hz OLED display. The faster speed lets you respond quickly to opponents and stay a…
【Powerful Connectivity】 2780M 2560 x 1440p 500hz gaming monitor delivers powerful connectivity: 2 x DP, 2 x HDMI, 1 x 3.5MM Audio Jack, wide…
The display itself is also quite forgiving in terms of placement, since the special anti-glare coating does an excellent job at maintaining visibility regardless of the lighting conditions in your room. With that level of control over reflections, your emphasis remains where it should be: on the game. For the competitive crowd, there is an extra tiny tool in the arsenal known as tournament mode. When you turn it on, the display basically shrinks to 24.5 inches, with black bars on the sides, but you still get the desired high refresh rate.
Ergonomically, the setup feels perfectly natural on almost any workstation. The stand can tilt from minus five to thirty-five degrees and adjusts in height by roughly five inches to maintain your screen at the ideal angle. Plus, it swivels left and right, allowing you to have a good perspective regardless of your preferences.
In terms of input, you have two HDMI 2.1 connections and one DisplayPort 2.1 connector to keep up with the latest graphics cards. Variable refresh rate support almost guarantees that you’ll never have to struggle with those annoying screen tearing bugs. As an added bonus, you get two pre-tuned picture settings for shooter games: one that gives you the familiar look of a regular display, and another that really shows off the OLED panel. Sony plans to sell the monitor for $1,099, with a release later this year. [Source]
You must be logged in to post a comment Login