Frontier AI companies often refer to their models as “helpful assistants” or try to compare them to entry-level employees.  

But new research from the UK’s AI Security Institute reinforces how large language models suffer from a common flaw that would land many human employees in hot water with their employers: they cheat.

In other words, these models are so committed to completing their tasks that they will break the rules, cut corners and deceive  their own users to accomplish them.

“Every model we have tested for this behavior attempted to cheat,” the AISI report said.
“Models did not reliably report this behavior when asked, and often did not reason about it in their chain-of-thought, suggesting that detecting cheating will likely require robust monitoring methods.”

The research tested OpenAI’s ChatGPT 5.4, 5.5 and 5.6 models, along with Anthropic’s Claude Opus 4.7 and Mythos Preview.

The AISI ran models through a series of “Capture-the-Flag” cyber evaluations, where they obtained points by performing offensive cybersecurity related tasks, like exploiting a particular vulnerability or reverse engineering compiled code.

The research defines “cheating” as “taking an action that is out of scope for the task or explicitly disallowed by the rules, in order to achieve a goal through a shortcut, workaround, or unintended solution that the task was not meant to, or should not, permit.”

It captures a range of observed behaviors from models, like searching the internet for solutions, attacking or escalating privileges on unrelated systems, probing evaluation software to gain access to the task solution.

Models also failed to acknowledge when they were cheating to complete a task, and less than 50 percent said the rule-breaking was “wrong” when challenged on it by a user.

A model’s propensity for cheating was not related to its capability, meaning newer or advanced models aren’t more likely to cheat. Rather, researchers think it stems from the techniques used during a model’s training and alignment that are to blame.

But even if rates of cheating remain constant, the problem could still worsen over time. As newer models in the future could become more proficient and learn more effective cheating techniques.

This deception also makes it difficult for labs like AISI to verify their own work, which relies on evaluating trustworthy outputs from AI systems.

The research underscores how AI systems can go to drastic lengths to complete their task, including blowing through or circumventing a company’s IT and cybersecurity protections.

In one instance, AISI researchers said a model was inadvertently given a cyber capability evaluation that was misconfigured and impossible to solve.

“The model tested was so persistent in attempting to cheat that it wrote and ran code on an external service, hosted on the open internet outside of AISI’s systems, in an attempt to access our evaluation infrastructure, triggering a security alert in AISI’s systems,” the report said.

While AISI said there were no data leaks or damage from the incident, the model could have successfully accessed their evaluation system had they not had monitoring in place. The institute said it implemented further controls on internal systems in response to the test.

The researchers said there are “significant consequences” to a status quo where we can’t trust models not to cheat. The behaviors are especially problematic in areas like AI safety and security research, as well as cyber operations and military decision-making, where trust outputs from the AI systems are critical.

Today, AISI said it can detect LLM cheating through a mix of manual review and LLM monitoring, but that may not always be true, and future models may be better at hiding their actions from human overseers.

“A more fundamental fix would be to train the models not to cheat in the first place – but given this kind of behavior was reported in frontier models more than a year ago, robustly aligning it away may not be easy,” researchers wrote.

The post AI models keep getting caught cheating appeared first on CyberScoop.

Reuters reports: Incoming British prime minister Andy Burnham will scrap the government's troubled plans for a digital ID scheme when he enters office on Monday, a spokesperson for the new Labour Party leader said. Resources devoted to the scheme, deemed a "fiasco" by a cross-party committee of lawmakers, will be redirected to Burnham's priorities, the spokesperson said... "All the time and resource that was going to be spent on a national ID scheme will go instead to where it's most needed, such as helping with the cost of living," Burnham's spokesperson said. In November, the Office for Budget Responsibility watchdog estimated the cost of the digital ID scheme at around £1.8 billion ($2.4 billion) between financial years 2026/27 and 2028/29.

Read more of this story at Slashdot.

A pair of young men were sentenced to 66 months in jail for committing a cyberattack on the Transport for London that brought the network’s operations to a standstill in 2024, the United Kingdom’s National Crime Agency said Thursday.

Thalha Jubair and Owen Flowers were arrested at their homes in September 2025, barely a year after the attack, and pleaded guilty last month just as their trials were set to begin. Flowers was previously arrested in connection with the attack in September, but was released after questioning by officers.

Jubair and Flowers were leading members and highly involved in Scattered Spider, a nebulous hacker subset of The Com, according to researchers. The 20-year-old Jubair was a prolific cybercriminal and core member of the unbound collective. 

U.S. authorities last year accused Jubair of direct, prominent involvement in at least 120 cyberattacks, including extortion of 47 U.S.-based organizations and the January 2025 attack on the federal court system. 

Officials said they traced a combined total of at least $89.5 million in cryptocurrency, at the time of payments, to Bitcoin addresses and servers controlled by Jubair. Two financial services firms paid Jubair $25 million and $36.2 million, respectively, in Bitcoin between June and November 2023, according to an unsealed criminal complaint against Jubair. 

At the time of Jubair’s arrest, “he was one of the four principal people that we associated with Scattered Spider,” and one of the two most core players, Adam Meyers, senior vice president of counter adversary operations at CrowdStrike, told CyberScoop. 

Jubair and Owens had significant resources and support, and “victim payments were reinvested back into the enterprise,” said Allison Nixon, chief research officer at Unit 221B. 

The lasting impact of Jubair and Owens’ capture and imprisonment remains hazy.

U.K. authorities insist Jubair and Owens’ arrests and punishment “effectively halted the group’s criminal activity,” yet they added that other cybercriminals continue to use the Scattered Spider brand in more recent attacks. 

Thursday’s announcement “represents a significant step in holding accountable two members of Scattered Spider, a group that has repeatedly relied on data extortion, SIM-swap attacks, and other social engineering techniques to infiltrate networks and undermine critical services,” Brett Leatherman, assistant director of the FBI Cyber Division, said in a statement. 

The FBI also noted, in a LinkedIn post, that members of Scattered Spider “continue to victimize organizations around the world and cause significant financial and operational harm.”

When Owens, now 18, was first arrested for the Transport for London attack in 2024, investigators said he was “in the process of hacking the systems of U.S. health care companies SSM Health Care Corporation and Sutter Health, which had been infiltrated and damaged.”

Officials also said Jubair and Owens failed to cooperate after their arrests. 

“This is the largest cybercrime prosecution ever brought before the U.K. courts and the culmination of nearly two years of painstaking work,” Paul Foster, head of the National Crime Center’s National Cybercrime Unit, said in a statement. 

“Scattered Spider has been the most significant cybercrime threat to the U.K. in recent years. Through this investigation, we have severely disrupted that threat and brought key offenders to justice,” Foster added.

Despite the upbeat reaction from U.K. officials, Nixon said the punishment for Jubair and Owens is “remarkably lenient considering the period of continuous reoffending lasted longer than the sentence.”

Nixon hopes the United States will eventually extradite the pair to face additional charges. “If that happens, they won’t be able to use mental illness as a loophole to get back to harming society as soon as possible,” she added.

“No one who worked on their case was surprised they would reoffend, and there seems to be no allowance in the law to protect the public from what everyone knew was going to happen,” Nixon said. “I know the narrative in the cybercriminal culture will glorify them, but they wouldn’t if they knew the full story.”

The post Leading members of Scattered Spider sentenced in UK to 66 months in jail appeared first on CyberScoop.

Thalha Jubair and Owen Flowers were prosecuted over a 2024 cyberattack targeting Transport for London (TfL).

The post Two Scattered Spider Hackers Sentenced to Jail in UK appeared first on SecurityWeek.

Facial recognition technology in U.K. shops "will soon alert police in real time to the presence of serious offenders," reports The Guardian, "with civil liberties groups warning of a 'dangerous escalation' towards surveillance and criminalisation in the retail sector." Facewatch, a facial recognition system used by more than 100 businesses including Sainsbury's, B&M and Spar to monitor thieves, said it was launching a UK-first feature to "alert police instantly when the most serious offenders trigger a live facial recognition match". Facewatch's chief executive, Nick Fisher, said the "unique technical development" would be launched in autumn and would warn police in an average of four seconds when the "worst offenders" were flagged on its network... Charlie Whelton, the policy and campaigns officer at [civil liberties nonprofit] Liberty, said it was concerned about this "untested, opaque development" and the way facial recognition technology had been allowed to "proliferate without anything to govern it". "It's not against the law to walk into a shop even if you've committed crimes in the past," he said. "The idea of calling the police on somebody who hasn't committed a crime, but there's a concern they might, is really upending the way we do things. And of course, it's not infallible. These systems do make mistakes, and it's very hard to argue with that when it happens to you." A number of people have been forced to leave shops after being falsely identified by Facewatch technology as a shoplifter, with some describing it as "Orwellian" and saying they felt as though they were "guilty until proven innocent"... The use of the Facewatch technology looks set to quickly expand, with Sainsbury's recently announcing plans to increase its use from 55 stores to more than 200 by the end of the year. Facewatch said it alerted retailers almost 300,000 times that a "known repeat offender" had entered a store during the first six months of 2026, and that its system allowed staff to intervene "before theft, abuse or violence could occur or escalate"... [E]xperts argue the use of facial recognition technology in shops to catch shoplifters is disproportionate. Nuala Polo, the UK public policy lead at the Ada Lovelace Institute, which studies the impact of AI on society, said: "There are other, much less intrusive means that you can use to catch shoplifters where you don't need to be scanning millions of faces every day, virtually without consent...." The campaign group Big Brother Watch has criticised police for "inserting themselves into this cowboy operation" and said people would be matched against "a secret blacklist compiled by unaccountable businesses and private security guards".

Read more of this story at Slashdot.

An Armenian national who was extradited from Ukraine to the United States last year pleaded guilty to participating in a series of attacks in 2019 and 2020 involving Ryuk ransomware, the Justice Department said Thursday.

Karen Serobovich Vardanyan pleaded guilty to computer fraud and conspiracy to commit fraud and extortion. He agreed to pay nearly $1.2 million million in restitution and faces up to 15 years in jail.

The 34-year-old admitted to participating in cybercrime from November 2019 to April 2020 when he and his co-conspirators deployed Ryuk ransomware against three U.S.-based organizations while living in Ukraine and Russia.

Vardanyan’s victims include a Michigan-based company that paid a ransom of nearly $1.2 million in January 2020, a Watsonville, Oregon-based technology company that was attacked in December 2019 and a Texas-based school breached in February 2020.

Prosecutors previously accused Vardanyan and his co-conspirators — Ukrainian nationals Oleg Nikolayevich Lyulyava and Andrii Leonydovich Prykhodchenko, and Armenian national Levon Georgiyovych Avetisyan — of illegally accessing computer networks to deploy Ryuk ransomware on hundreds of compromised servers and workstations between March 2019 and September 2020.

Ryuk ransomware was prevalent in 2019 and 2020, infecting thousands of victims globally across the private sector, state and local municipalities, local school districts and critical infrastructure, including a wave of attacks on U.S. hospitals.

Victims of Ryuk ransomware attacks include Hollywood Presbyterian Medical Center, Universal Health Services, Electronic Warfare Associates, a North Carolina water utility and multiple U.S. newspapers.

Ryuk ransomware operators extorted victim companies by demanding ransom payments in Bitcoin in exchange for decryption keys. Justice Department officials said Vardanyan and his co-conspirators received about 1,160 bitcoins — valued at more than $15 million at the time — in ransom payments from victim companies.

Vardanyan, as part of his guilty plea, also acknowledged that his conviction will have immigration consequences resulting in removal from the United States after serving his sentence. 

The U.S. District Court for the District of Oregon has yet to schedule his sentencing.

The post Armenian national pleads guilty to Ryuk ransomware attacks appeared first on CyberScoop.

Two announcements on July 7, 2026, demonstrate the government’s determination to improve the level of cybersecurity within the UK.

The post UK Government Rolls Out Agentic AI Defense Plan Alongside Industry Pledge appeared first on SecurityWeek.

A private tutor who charged money to take dozens of exams for students and submit coursework for them "has been jailed for three years," reports the BBC, "after his scam earned him £300,000." Shahid Adnan completed assignments and online tests for more than 120 students at Liverpool John Moore's University, the Crown Prosecution Service said. The 43-year-old, of Lysander Close, Liverpool, was caught in February 2023 after a student handed in a USB drive containing suspicious coursework to Dr Tom Berry of the university's school of computer science and mathematics. Berry's checks revealed the drive was used by Adnan with documents linked to a company he set up called Study Sharp Ltd. Excel spreadsheets containing details of other students, their study modules, coursework due dates, and their personal login credentials were also found. Further checks confirmed suspicions that Adnan was accessing the university's network to submit fraudulent work and sit examinations on behalf of students... [I]nvestigations led police to believe Adnan may have been doing work for 124 students at universities all over the world. The BBC also interviewed detective sergeant Adam Dagnall from Merseyside Police's cybercrime unit, who said Adnan was living a lavish lifestyle "well beyond" his stated occupations as a private tutor and Amazon delivery driver. His bank accounts held more than £2m ($2,645,100 USD).

Read more of this story at Slashdot.

PC Gamer reports: The UK government is considering an Australia-style ban on social media for under-16s, with Prime Minister Keir Starmer saying that the ban could take effect as soon as spring next year. As for the much nearer future, Science and Technology Secretary Liz Kendall told BBC Breakfast earlier this week, "We will make further statements in July about VPNs and further restrictions." To be clear, no specific restrictions have yet been announced and Kendall sounded somewhat cautious about an outright ban during a parliament debate that took place the same day. "I have commissioned further research about their usage. There are really important issues to balance here," she says. "Many people want to use VPNs for privacy — that is important — but we know that some children use them to get around restrictions. I will come back to that in July in our response to the consultation." So, we'll have to wait until next month for anything definite, but it's hard not to feel like a full ban on VPNs is already on the table. If that does come to pass, more than the contents of my Bluesky inbox will be at stake. Utah in the US has already tried to implement a full VPN ban (though this was postponed until September after Aylo, the parent company of Pornhub, challenged the law in court)... [T]he UK could just be the next domino after Utah, potentially setting off a chain reaction that affects users around the world. The article also argues that age checks can also be a privacy nightmare "with the security breach that exposed the personal info of 70,000 Discord users last year being one case in point." Here's the complete statement from UK Technology Secretary Kendall. "I'll come back in July with a further statement around VPNs but also additional measures that we want to look at, further restrictions on AI chatbots that parents have found very worrying, more about overnight curfews or breaks in doomscrolling for 16- and 17-year-olds."

Read more of this story at Slashdot.

The Sunday Times reports: A criminal investigation has begun after a police officer allegedly used AI to create evidential material in a "number of cases". Derbyshire Constabulary said an officer was being investigated over an allegation of suspected perverting the course of justice. The Crown Prosecution Service (CPS) confirmed it was engaging with defence lawyers and the courts over potentially affected cases... It is the first known allegation of AI misuse by police in a criminal case in the UK, but it follows an incident last year in which West Midlands police relied on AI-generated material that fabricated a match involving Maccabi Tel Aviv. The material was used in intelligence supporting a proposed ban on away fans at the club's match against Aston Villa.

Read more of this story at Slashdot.

An elusive hacker who went by the handle “UNKN” and ran the early Russian ransomware groups GandCrab and REvil now has a name and a face. Authorities in Germany say 31-year-old Russian Daniil Maksimovich Shchukin headed both cybercrime gangs and helped carry out at least 130 acts of computer sabotage and extortion against victims across the country between 2019 and 2021.

Shchukin was named as UNKN (a.k.a. UNKNOWN) in an advisory published by the German Federal Criminal Police (the “Bundeskriminalamt” or BKA for short). The BKA said Shchukin and another Russian — 43-year-old Anatoly Sergeevitsch Kravchuk — extorted nearly $2 million euros across two dozen cyberattacks that caused more than 35 million euros in total economic damage.

Germany’s BKA said Shchukin acted as the head of one of the largest worldwide operating ransomware groups GandCrab and REvil, which pioneered the practice of double extortion — charging victims once for a key needed to unlock hacked systems, and a separate payment in exchange for a promise not to publish stolen data.

Shchukin’s name appeared in a Feb. 2023 filing (PDF) from the U.S. Justice Department seeking the seizure of various cryptocurrency accounts associated with proceeds from the REvil ransomware gang’s activities. The government said the digital wallet tied to Shchukin contained more than $317,000 in ill-gotten cryptocurrency.

The GandCrab ransomware affiliate program first surfaced in January 2018, and paid enterprising hackers huge shares of the profits just for hacking into user accounts at major corporations. The GandCrab team would then try to expand that access, often siphoning vast amounts of sensitive and internal documents in the process. The malware’s curators shipped five major revisions to the GandCrab code, each corresponding with sneaky new features and bug fixes aimed at thwarting the efforts of computer security firms to stymie the spread of the malware.

On May 31, 2019, the GandCrab team announced the group was shutting down after extorting more than $2 billion from victims. “We are a living proof that you can do evil and get off scot-free,” GandCrab’s farewell address famously quipped. “We have proved that one can make a lifetime of money in one year. We have proved that you can become number one by general admission, not in your own conceit.”

The REvil ransomware affiliate program materialized around the same as GandCrab’s demise, fronted by a user named UNKNOWN who announced on a Russian cybercrime forum that he’d deposited $1 million in the forum’s escrow to show he meant business. By this time, many cybersecurity experts had concluded REvil was little more than a reorganization of GandCrab.

UNKNOWN also gave an interview to Dmitry Smilyanets, a former malicious hacker hired by Recorded Future, wherein UNKNOWN described a rags-to-riches tale unencumbered by ethics and morals.

“As a child, I scrounged through the trash heaps and smoked cigarette butts,” UNKNOWN told Recorded Future. “I walked 10 km one way to the school. I wore the same clothes for six months. In my youth, in a communal apartment, I didn’t eat for two or even three days. Now I am a millionaire.”

As described in The Ransomware Hunting Team by Renee Dudley and Daniel Golden, UNKNOWN and REvil reinvested significant earnings into improving their success and mirroring practices of legitimate businesses. The authors wrote:

“Just as a real-world manufacturer might hire other companies to handle logistics or web design, ransomware developers increasingly outsourced tasks beyond their purview, focusing instead on improving the quality of their ransomware. The higher quality ransomware—which, in many cases, the Hunting Team could not break—resulted in more and higher pay-outs from victims. The monumental payments enabled gangs to reinvest in their enterprises. They hired more specialists, and their success accelerated.”

“Criminals raced to join the booming ransomware economy. Underworld ancillary service providers sprouted or pivoted from other criminal work to meet developers’ demand for customized support. Partnering with gangs like GandCrab, ‘cryptor’ providers ensured ransomware could not be detected by standard anti-malware scanners. ‘Initial access brokerages’ specialized in stealing credentials and finding vulnerabilities in target networks, selling that access to ransomware operators and affiliates. Bitcoin “tumblers” offered discounts to gangs that used them as a preferred vendor for laundering ransom payments. Some contractors were open to working with any gang, while others entered exclusive partnerships.”

REvil would evolve into a feared “big-game-hunting” machine capable of extracting hefty extortion payments from victims, largely going after organizations with more than $100 million in annual revenues and fat new cyber insurance policies that were known to pay out.

Over the July 4, 2021 weekend in the United States, REvil hacked into and extorted Kaseya, a company that handled IT operations for more than 1,500 businesses, nonprofits and government agencies. The FBI would later announce they’d infiltrated the ransomware group’s servers prior to the Kaseya hack but couldn’t tip their hand at the time. REvil never recovered from that core compromise, or from the FBI’s release of a free decryption key for REvil victims who couldn’t or didn’t pay.

Shchukin is from Krasnodar, Russia and is thought to reside there, the BKA said.

“Based on the investigations so far, it is assumed that the wanted person is abroad, presumably in Russia,” the BKA advised. “Travel behaviour cannot be ruled out.”

There is little that connects Shchukin to UNKNOWN’s various accounts on the Russian crime forums. But a review of the Russian crime forums indexed by the cyber intelligence firm Intel 471 shows there is plenty connecting Shchukin to a hacker identity called “Ger0in” who operated large botnets and sold “installs” — allowing other cybercriminals to rapidly deploy malware of their choice to thousands of PCs in one go. However, Ger0in was only active between 2010 and 2011, well before UNKNOWN’s appearance as the REvil front man.

A review of the mugshots released by the BKA at the image comparison site Pimeyes found a match on this birthday celebration from 2023, which features a young man named Daniel wearing the same fancy watch as in the BKA photos.

Update, April 6, 12:06 p.m. ET: A reader forwarded this English-dubbed audio recording from a ccc.de (37C3) conference talk in Germany from 2023 that previously outed Shchukin as the REvil leader (Shchuckin is mentioned at around 24:25).

Microsoft today pushed updates to fix at least 56 security flaws in its Windows operating systems and supported software. This final Patch Tuesday of 2025 tackles one zero-day bug that is already being exploited, as well as two publicly disclosed vulnerabilities.

Despite releasing a lower-than-normal number of security updates these past few months, Microsoft patched a whopping 1,129 vulnerabilities in 2025, an 11.9% increase from 2024. According to Satnam Narang at Tenable, this year marks the second consecutive year that Microsoft patched over one thousand vulnerabilities, and the third time it has done so since its inception.

The zero-day flaw patched today is CVE-2025-62221, a privilege escalation vulnerability affecting Windows 10 and later editions. The weakness resides in a component called the “Windows Cloud Files Mini Filter Driver” — a system driver that enables cloud applications to access file system functionalities.

“This is particularly concerning, as the mini filter is integral to services like OneDrive, Google Drive, and iCloud, and remains a core Windows component, even if none of those apps were installed,” said Adam Barnett, lead software engineer at Rapid7.

Only three of the flaws patched today earned Microsoft’s most-dire “critical” rating: Both CVE-2025-62554 and CVE-2025-62557 involve Microsoft Office, and both can exploited merely by viewing a booby-trapped email message in the Preview Pane. Another critical bug — CVE-2025-62562 — involves Microsoft Outlook, although Redmond says the Preview Pane is not an attack vector with this one.

But according to Microsoft, the vulnerabilities most likely to be exploited from this month’s patch batch are other (non-critical) privilege escalation bugs, including:

–CVE-2025-62458 — Win32k
–CVE-2025-62470 — Windows Common Log File System Driver
–CVE-2025-62472 — Windows Remote Access Connection Manager
–CVE-2025-59516 — Windows Storage VSP Driver
–CVE-2025-59517 — Windows Storage VSP Driver

Kev Breen, senior director of threat research at Immersive, said privilege escalation flaws are observed in almost every incident involving host compromises.

“We don’t know why Microsoft has marked these specifically as more likely, but the majority of these components have historically been exploited in the wild or have enough technical detail on previous CVEs that it would be easier for threat actors to weaponize these,” Breen said. “Either way, while not actively being exploited, these should be patched sooner rather than later.”

One of the more interesting vulnerabilities patched this month is CVE-2025-64671, a remote code execution flaw in the Github Copilot Plugin for Jetbrains AI-based coding assistant that is used by Microsoft and GitHub. Breen said this flaw would allow attackers to execute arbitrary code by tricking the large language model (LLM) into running commands that bypass the user’s “auto-approve” settings.

CVE-2025-64671 is part of a broader, more systemic security crisis that security researcher Ari Marzuk has branded IDEsaster (IDE  stands for “integrated development environment”), which encompasses more than 30 separate vulnerabilities reported in nearly a dozen market-leading AI coding platforms, including Cursor, Windsurf, Gemini CLI, and Claude Code.

The other publicly-disclosed vulnerability patched today is CVE-2025-54100, a remote code execution bug in Windows Powershell on Windows Server 2008 and later that allows an unauthenticated attacker to run code in the security context of the user.

For anyone seeking a more granular breakdown of the security updates Microsoft pushed today, check out the roundup at the SANS Internet Storm Center. As always, please leave a note in the comments if you experience problems applying any of this month’s Windows patches.