Marko Elez, a 25-year-old employee at Elon Musk’s Department of Government Efficiency (DOGE), has been granted access to sensitive databases at the U.S. Social Security Administration, the Treasury and Justice departments, and the Department of Homeland Security. So it should fill all Americans with a deep sense of confidence to learn that Mr. Elez over the weekend inadvertently published a private key that allowed anyone to interact directly with more than four dozen large language models (LLMs) developed by Musk’s artificial intelligence company xAI.

On July 13, Mr. Elez committed a code script to GitHub called “agent.py” that included a private application programming interface (API) key for xAI. The inclusion of the private key was first flagged by GitGuardian, a company that specializes in detecting and remediating exposed secrets in public and proprietary environments. GitGuardian’s systems constantly scan GitHub and other code repositories for exposed API keys, and fire off automated alerts to affected users.

Philippe Caturegli, “chief hacking officer” at the security consultancy Seralys, said the exposed API key allowed access to at least 52 different LLMs used by xAI. The most recent LLM in the list was called “grok-4-0709” and was created on July 9, 2025.

Grok, the generative AI chatbot developed by xAI and integrated into Twitter/X, relies on these and other LLMs (a query to Grok before publication shows Grok currently uses Grok-3, which was launched in Feburary 2025). Earlier today, xAI announced that the Department of Defense will begin using Grok as part of a contract worth up to $200 million. The contract award came less than a week after Grok began spewing antisemitic rants and invoking Adolf Hitler.

Mr. Elez did not respond to a request for comment. The code repository containing the private xAI key was removed shortly after Caturegli notified Elez via email. However, Caturegli said the exposed API key still works and has not yet been revoked.

“If a developer can’t keep an API key private, it raises questions about how they’re handling far more sensitive government information behind closed doors,” Caturegli told KrebsOnSecurity.

Prior to joining DOGE, Marko Elez worked for a number of Musk’s companies. His DOGE career began at the Department of the Treasury, and a legal battle over DOGE’s access to Treasury databases showed Elez was sending unencrypted personal information in violation of the agency’s policies.

While still at Treasury, Elez resigned after The Wall Street Journal linked him to social media posts that advocated racism and eugenics. When Vice President J.D. Vance lobbied for Elez to be rehired, President Trump agreed and Musk reinstated him.

Since his re-hiring as a DOGE employee, Elez has been granted access to databases at one federal agency after another. TechCrunch reported in February 2025 that he was working at the Social Security Administration. In March, Business Insider found Elez was part of a DOGE detachment assigned to the Department of Labor.

In April, The New York Times reported that Elez held positions at the U.S. Customs and Border Protection and the Immigration and Customs Enforcement (ICE) bureaus, as well as the Department of Homeland Security. The Washington Post later reported that Elez, while serving as a DOGE advisor at the Department of Justice, had gained access to the Executive Office for Immigration Review’s Courts and Appeals System (EACS).

Elez is not the first DOGE worker to publish internal API keys for xAI: In May, KrebsOnSecurity detailed how another DOGE employee leaked a private xAI key on GitHub for two months, exposing LLMs that were custom made for working with internal data from Musk’s companies, including SpaceX, Tesla and Twitter/X.

Caturegli said it’s difficult to trust someone with access to confidential government systems when they can’t even manage the basics of operational security.

“One leak is a mistake,” he said. “But when the same type of sensitive key gets exposed again and again, it’s not just bad luck, it’s a sign of deeper negligence and a broken security culture.”

Authorities in the United Kingdom this week arrested four alleged members of “Scattered Spider,” a prolific data theft and extortion group whose recent victims include multiple airlines and the U.K. retail chain Marks & Spencer.

Scattered Spider is the name given to an English-speaking cybercrime group known for using social engineering tactics to break into companies and steal data for ransom, often impersonating employees or contractors to deceive IT help desks into granting access. The FBI warned last month that Scattered Spider had recently shifted to targeting companies in the retail and airline sectors.

The U.K.’s National Crime Agency (NCA) declined verify the names of those arrested, saying only that they included two males aged 19, another aged 17, and 20-year-old female. The NCA said the defendants were charged in cyberattacks against Marks & Spencer, the U.K. retailer Harrods, and the British food retailer Co-op Group.

KrebsOnSecurity has learned the identities of two of the suspects. Multiple sources close to the investigation said those arrested include Owen David Flowers, a U.K. man alleged to have been involved in the cyber intrusion and ransomware attack that shut down several MGM Casino properties in September 2023. Those same sources said the woman arrested is or recently was in a relationship with Flowers.

Sources told KrebsOnSecurity that Flowers, who allegedly went by the hacker handles “bo764,” “Holy,” and “Nazi,” was the group member who anonymously gave interviews to the media in the days after the MGM hack. His real name was omitted from a September 2024 story about the group because he was not yet charged in that incident.

The bigger fish netted as part of the Scattered Spider dragnet is 19-year-old Thalha Jubair, a U.K. man whose alleged exploits under various monikers have been well-documented in stories on this site. Jubair is believed to have used the nickname “Earth2Star,” which corresponds to a founding member of the cybercrime-focused Telegram channel “Star Fraud Chat.”

In 2023, KrebsOnSecurity published an investigation into the work of three different SIM-swapping groups that phished credentials from T-Mobile employees and used that access to offer a service whereby any T-Mobile phone number could be swapped to a new device. Star Chat was by far the most active and consequential of the three SIM-swapping groups, who collectively broke into T-Mobile’s network more than 100 times in the second half of 2022.

Sources tell KrebsOnSecurity that Jubair also was a core member of the LAPSUS$ cybercrime group that broke into dozens of technology companies in 2022, stealing source code and other internal data from tech giants including Microsoft, Nvidia, Okta, Rockstar Games, Samsung, T-Mobile, and Uber.

In April 2022, KrebsOnSecurity published internal chat records from LAPSUS$, and those chats indicated Jubair was using the nicknames Amtrak and Asyntax. At one point in the chats, Amtrak told the LAPSUS$ group leader not to share T-Mobile’s logo in images sent to the group because he’d been previously busted for SIM-swapping and his parents would suspect he was back at it again.

As shown in those chats, the leader of LAPSUS$ eventually decided to betray Amtrak by posting his real name, phone number, and other hacker handles into a public chat room on Telegram.

That story about the leaked LAPSUS$ chats connected Amtrak/Asyntax/Jubair to the identity “Everlynn,” the founder of a cybercriminal service that sold fraudulent “emergency data requests” targeting the major social media and email providers. In such schemes, the hackers compromise email accounts tied to police departments and government agencies, and then send unauthorized demands for subscriber data while claiming the information being requested can’t wait for a court order because it relates to an urgent matter of life and death.

Sources say Jubair also used the nickname “Operator,” and that until recently he was the administrator of the Doxbin, a long-running and highly toxic online community that is used to “dox” or post deeply personal information on people. In May 2024, several popular cybercrime channels on Telegram ridiculed Operator after it was revealed that he’d staged his own kidnapping in a botched plan to throw off law enforcement investigators.

In November 2024, U.S. authorities charged five men aged 20 to 25 in connection with the Scattered Spider group, which has long relied on recruiting minors to carry out its most risky activities. Indeed, many of the group’s core members were recruited from online gaming platforms like Roblox and Minecraft in their early teens, and have been perfecting their social engineering tactics for years.

“There is a clear pattern that some of the most depraved threat actors first joined cybercrime gangs at an exceptionally young age,” said Allison Nixon, chief research officer at the New York based security firm Unit 221B. “Cybercriminals arrested at 15 or younger need serious intervention and monitoring to prevent a years long massive escalation.”

Microsoft today released updates to fix at least 137 security vulnerabilities in its Windows operating systems and supported software. None of the weaknesses addressed this month are known to be actively exploited, but 14 of the flaws earned Microsoft’s most-dire “critical” rating, meaning they could be exploited to seize control over vulnerable Windows PCs with little or no help from users.

While not listed as critical, CVE-2025-49719 is a publicly disclosed information disclosure vulnerability, with all versions as far back as SQL Server 2016 receiving patches. Microsoft rates CVE-2025-49719 as less likely to be exploited, but the availability of proof-of-concept code for this flaw means its patch should probably be a priority for affected enterprises.

Mike Walters, co-founder of Action1, said CVE-2025-49719 can be exploited without authentication, and that many third-party applications depend on SQL server and the affected drivers — potentially introducing a supply-chain risk that extends beyond direct SQL Server users.

“The potential exposure of sensitive information makes this a high-priority concern for organizations handling valuable or regulated data,” Walters said. “The comprehensive nature of the affected versions, spanning multiple SQL Server releases from 2016 through 2022, indicates a fundamental issue in how SQL Server handles memory management and input validation.”

Adam Barnett at Rapid7 notes that today is the end of the road for SQL Server 2012, meaning there will be no future security patches even for critical vulnerabilities, even if you’re willing to pay Microsoft for the privilege.

Barnett also called attention to CVE-2025-47981, a vulnerability with a CVSS score of 9.8 (10 being the worst), a remote code execution bug in the way Windows servers and clients negotiate to discover mutually supported authentication mechanisms. This pre-authentication vulnerability affects any Windows client machine running Windows 10 1607 or above, and all current versions of Windows Server. Microsoft considers it more likely that attackers will exploit this flaw.

Microsoft also patched at least four critical, remote code execution flaws in Office (CVE-2025-49695, CVE-2025-49696, CVE-2025-49697, CVE-2025-49702). The first two are both rated by Microsoft as having a higher likelihood of exploitation, do not require user interaction, and can be triggered through the Preview Pane.

Two more high severity bugs include CVE-2025-49740 (CVSS 8.8) and CVE-2025-47178 (CVSS 8.0); the former is a weakness that could allow malicious files to bypass screening by Microsoft Defender SmartScreen, a built-in feature of Windows that tries to block untrusted downloads and malicious sites.

CVE-2025-47178 involves a remote code execution flaw in Microsoft Configuration Manager, an enterprise tool for managing, deploying, and securing computers, servers, and devices across a network. Ben Hopkins at Immersive Labs said this bug requires very low privileges to exploit, and that it is possible for a user or attacker with a read-only access role to exploit it.

“Exploiting this vulnerability allows an attacker to execute arbitrary SQL queries as the privileged SMS service account in Microsoft Configuration Manager,” Hopkins said. “This access can be used to manipulate deployments, push malicious software or scripts to all managed devices, alter configurations, steal sensitive data, and potentially escalate to full operating system code execution across the enterprise, giving the attacker broad control over the entire IT environment.”

Separately, Adobe has released security updates for a broad range of software, including After Effects, Adobe Audition, Illustrator, FrameMaker, and ColdFusion.

The SANS Internet Storm Center has a breakdown of each individual patch, indexed by severity. If you’re responsible for administering a number of Windows systems, it may be worth keeping an eye on AskWoody for the lowdown on any potentially wonky updates (considering the large number of vulnerabilities and Windows components addressed this month).

If you’re a Windows home user, please consider backing up your data and/or drive before installing any patches, and drop a note in the comments if you encounter any problems with these updates.

In May 2025, the U.S. government sanctioned a Chinese national for operating a cloud provider linked to the majority of virtual currency investment scam websites reported to the FBI. But a new report finds the accused continues to operate a slew of established accounts at American tech companies — including Facebook, Github, PayPal and Twitter/X.

On May 29, the U.S. Department of the Treasury announced economic sanctions against Funnull Technology Inc., a Philippines-based company alleged to provide infrastructure for hundreds of thousands of websites involved in virtual currency investment scams known as “pig butchering.” In January 2025, KrebsOnSecurity detailed how Funnull was designed as a content delivery network that catered to foreign cybercriminals seeking to route their traffic through U.S.-based cloud providers.

The Treasury also sanctioned Funnull’s alleged operator, a 40-year-old Chinese national named Liu “Steve” Lizhi. The government says Funnull directly facilitated financial schemes resulting in more than $200 million in financial losses by Americans, and that the company’s operations were linked to the majority of pig butchering scams reported to the FBI.

It is generally illegal for U.S. companies or individuals to transact with people sanctioned by the Treasury. However, as Mr. Lizhi’s case makes clear, just because someone is sanctioned doesn’t necessarily mean big tech companies are going to suspend their online accounts.

The government says Lizhi was born November 13, 1984, and used the nicknames “XXL4” and “Nice Lizhi.” Nevertheless, Steve Liu’s 17-year-old account on LinkedIn (in the name “Liulizhi”) had hundreds of followers (Lizhi’s LinkedIn profile helpfully confirms his birthday) until quite recently: The account was deleted this morning, just hours after KrebsOnSecurity sought comment from LinkedIn.

In an emailed response, a LinkedIn spokesperson said the company’s “Prohibited countries policy” states that LinkedIn “does not sell, license, support or otherwise make available its Premium accounts or other paid products and services to individuals and companies sanctioned by the U.S. government.” LinkedIn declined to say whether the profile in question was a premium or free account.

Mr. Lizhi also maintains a working PayPal account under the name Liu Lizhi and username “@nicelizhi,” another nickname listed in the Treasury sanctions. PayPal did not respond to a request for comment. A 15-year-old Twitter/X account named “Lizhi” that links to Mr. Lizhi’s personal domain remains active, although it has few followers and hasn’t posted in years.

These accounts and many others were flagged by the security firm Silent Push, which has been tracking Funnull’s operations for the past year and calling out U.S. cloud providers like Amazon and Microsoft for failing to more quickly sever ties with the company.

In a report released today, Silent Push found Lizhi still operates numerous Facebook accounts and groups, including a private Facebook account under the name Liu Lizhi. Another active Facebook account clearly connected to Lizhi is a tourism page for Ganzhou, China called “EnjoyGanzhou” that was named in the Treasury Department sanctions.

“This guy is the technical administrator for the infrastructure that is hosting a majority of scams targeting people in the United States, and hundreds of millions have been lost based on the websites he’s been hosting,” said Zach Edwards, senior threat researcher at Silent Push. “It’s crazy that the vast majority of big tech companies haven’t done anything to cut ties with this guy.”

The FBI says it received nearly 150,000 complaints last year involving digital assets and $9.3 billion in losses — a 66 percent increase from the previous year. Investment scams were the top crypto-related crimes reported, with $5.8 billion in losses.

In a statement, a Meta spokesperson said the company continuously takes steps to meet its legal obligations, but that sanctions laws are complex and varied. They explained that sanctions are often targeted in nature and don’t always prohibit people from having a presence on its platform. Nevertheless, Meta confirmed it had removed the account, unpublished Pages, and removed Groups and events associated with the user for violating its policies.

Attempts to reach Mr. Lizhi via his primary email addresses at Hotmail and Gmail bounced as undeliverable. Likewise, his 14-year-old YouTube channel appears to have been taken down recently.

However, anyone interested in viewing or using Mr. Lizhi’s 146 computer code repositories will have no problem finding active GitHub accounts for him, including one registered under the NiceLizhi and XXL4 nicknames mentioned in the Treasury sanctions.

Mr. Lizhi also operates a GitHub page for an open source e-commerce platform called NexaMerchant, which advertises itself as a payment gateway working with numerous American financial institutions. Interestingly, this profile’s “followers” page shows several other accounts that appear to be Mr. Lizhi’s. All of the account’s followers are tagged as “suspended,” even though that suspended message does not display when one visits those individual profiles.

In response to questions, GitHub said it has a process in place to identify when users and customers are Specially Designated Nationals or other denied or blocked parties, but that it locks those accounts instead of removing them. According to its policy, GitHub takes care that users and customers aren’t impacted beyond what is required by law.

“This includes keeping public repositories, including those for open source projects, available and accessible to support personal communications involving developers in sanctioned regions,” the policy states. “This also means GitHub will advocate for developers in sanctioned regions to enjoy greater access to the platform and full access to the global open source community.”

Edwards said it’s great that GitHub has a process for handling sanctioned accounts, but that the process doesn’t seem to communicate risk in a transparent way, noting that the only indicator on the locked accounts is the message, “This repository has been archived by the owner. It is not read-only.”

“It’s an odd message that doesn’t communicate, ‘This is a sanctioned entity, don’t fork this code or use it in a production environment’,” Edwards said.

Mark Rasch is a former federal cybercrime prosecutor who now serves as counsel for the New York City based security consulting firm Unit 221B. Rasch said when Treasury’s Office of Foreign Assets Control (OFAC) sanctions a person or entity, it then becomes illegal for businesses or organizations to transact with the sanctioned party.

Rasch said financial institutions have very mature systems for severing accounts tied to people who become subject to OFAC sanctions, but that tech companies may be far less proactive — particularly with free accounts.

“Banks have established ways of checking [U.S. government sanctions lists] for sanctioned entities, but tech companies don’t necessarily do a good job with that, especially for services that you can just click and sign up for,” Rasch said. “It’s potentially a risk and liability for the tech companies involved, but only to the extent OFAC is willing to enforce it.”

In July 2024, Funnull purchased the domain polyfill[.]io, the longtime home of a legitimate open source project that allowed websites to ensure that devices using legacy browsers could still render content in newer formats. After the Polyfill domain changed hands, at least 384,000 websites were caught in a supply-chain attack that redirected visitors to malicious sites. According to the Treasury, Funnull used the code to redirect people to scam websites and online gambling sites, some of which were linked to Chinese criminal money laundering operations.

The U.S. government says Funnull provides domain names for websites on its purchased IP addresses, using domain generation algorithms (DGAs) — programs that generate large numbers of similar but unique names for websites — and that it sells web design templates to cybercriminals.

“These services not only make it easier for cybercriminals to impersonate trusted brands when creating scam websites, but also allow them to quickly change to different domain names and IP addresses when legitimate providers attempt to take the websites down,” reads a Treasury statement.

Meanwhile, Funnull appears to be morphing nearly all aspects of its business in the wake of the sanctions, Edwards said.

“Whereas before they might have used 60 DGA domains to hide and bounce their traffic, we’re seeing far more now,” he said. “They’re trying to make their infrastructure harder to track and more complicated, so for now they’re not going away but more just changing what they’re doing. And a lot more organizations should be holding their feet to the fire.”

Update, 2:48 PM ET: Added response from Meta, which confirmed it has closed the accounts and groups connected to Mr. Lizhi.

Agents with the Federal Bureau of Investigation (FBI) briefed Capitol Hill staff recently on hardening the security of their mobile devices, after a contacts list stolen from the personal phone of the White House Chief of Staff Susie Wiles was reportedly used to fuel a series of text messages and phone calls impersonating her to U.S. lawmakers. But in a letter this week to the FBI, one of the Senate’s most tech-savvy lawmakers says the feds aren’t doing enough to recommend more appropriate security protections that are already built into most consumer mobile devices.

On May 29, The Wall Street Journal reported that federal authorities were investigating a clandestine effort to impersonate Ms. Wiles via text messages and in phone calls that may have used AI to spoof her voice. According to The Journal, Wiles told associates her cellphone contacts were hacked, giving the impersonator access to the private phone numbers of some of the country’s most influential people.

The execution of this phishing and impersonation campaign — whatever its goals may have been — suggested the attackers were financially motivated, and not particularly sophisticated.

“It became clear to some of the lawmakers that the requests were suspicious when the impersonator began asking questions about Trump that Wiles should have known the answers to—and in one case, when the impersonator asked for a cash transfer, some of the people said,” the Journal wrote. “In many cases, the impersonator’s grammar was broken and the messages were more formal than the way Wiles typically communicates, people who have received the messages said. The calls and text messages also didn’t come from Wiles’s phone number.”

Sophisticated or not, the impersonation campaign was soon punctuated by the murder of Minnesota House of Representatives Speaker Emerita Melissa Hortman and her husband, and the shooting of Minnesota State Senator John Hoffman and his wife. So when FBI agents offered in mid-June to brief U.S. Senate staff on mobile threats, more than 140 staffers took them up on that invitation (a remarkably high number considering that no food was offered at the event).

But according to Sen. Ron Wyden (D-Ore.), the advice the FBI provided to Senate staffers was largely limited to remedial tips, such as not clicking on suspicious links or attachments, not using public wifi networks, turning off bluetooth, keeping phone software up to date, and rebooting regularly.

“This is insufficient to protect Senate employees and other high-value targets against foreign spies using advanced cyber tools,” Wyden wrote in a letter sent today to FBI Director Kash Patel. “Well-funded foreign intelligence agencies do not have to rely on phishing messages and malicious attachments to infect unsuspecting victims with spyware. Cyber mercenary companies sell their government customers advanced ‘zero-click’ capabilities to deliver spyware that do not require any action by the victim.”

Wyden stressed that to help counter sophisticated attacks, the FBI should be encouraging lawmakers and their staff to enable anti-spyware defenses that are built into Apple’s iOS and Google’s Android phone software.

These include Apple’s Lockdown Mode, which is designed for users who are worried they may be subject to targeted attacks. Lockdown Mode restricts non-essential iOS features to reduce the device’s overall attack surface. Google Android devices carry a similar feature called Advanced Protection Mode.

Wyden also urged the FBI to update its training to recommend a number of other steps that people can take to make their mobile devices less trackable, including the use of ad blockers to guard against malicious advertisements, disabling ad tracking IDs in mobile devices, and opting out of commercial data brokers (the suspect charged in the Minnesota shootings reportedly used multiple people-search services to find the home addresses of his targets).

The senator’s letter notes that while the FBI has recommended all of the above precautions in various advisories issued over the years, the advice the agency is giving now to the nation’s leaders needs to be more comprehensive, actionable and urgent.

“In spite of the seriousness of the threat, the FBI has yet to provide effective defensive guidance,” Wyden said.

Nicholas Weaver is a researcher with the International Computer Science Institute, a nonprofit in Berkeley, Calif. Weaver said Lockdown Mode or Advanced Protection will mitigate many vulnerabilities, and should be the default setting for all members of Congress and their staff.

“Lawmakers are at exceptional risk and need to be exceptionally protected,” Weaver said. “Their computers should be locked down and well administered, etc. And the same applies to staffers.”

Weaver noted that Apple’s Lockdown Mode has a track record of blocking zero-day attacks on iOS applications; in September 2023, Citizen Lab documented how Lockdown Mode foiled a zero-click flaw capable of installing spyware on iOS devices without any interaction from the victim.

Earlier this month, Citizen Lab researchers documented a zero-click attack used to infect the iOS devices of two journalists with Paragon’s Graphite spyware. The vulnerability could be exploited merely by sending the target a booby-trapped media file delivered via iMessage. Apple also recently updated its advisory for the zero-click flaw (CVE-2025-43200), noting that it was mitigated as of iOS 18.3.1, which was released in February 2025.

Apple has not commented on whether CVE-2025-43200 could be exploited on devices with Lockdown Mode turned on. But HelpNetSecurity observed that at the same time Apple addressed CVE-2025-43200 back in February, the company fixed another vulnerability flagged by Citizen Lab researcher Bill Marczak: CVE-2025-24200, which Apple said was used in an extremely sophisticated physical attack against specific targeted individuals that allowed attackers to disable USB Restricted Mode on a locked device.

In other words, the flaw could apparently be exploited only if the attacker had physical access to the targeted vulnerable device. And as the old infosec industry adage goes, if an adversary has physical access to your device, it’s most likely not your device anymore.

I can’t speak to Google’s Advanced Protection Mode personally, because I don’t use Google or Android devices. But I have had Apple’s Lockdown Mode enabled on all of my Apple devices since it was first made available in September 2022. I can only think of a single occasion when one of my apps failed to work properly with Lockdown Mode turned on, and in that case I was able to add a temporary exception for that app in Lockdown Mode’s settings.

My main gripe with Lockdown Mode was captured in a March 2025 column by TechCrunch’s Lorenzo Francheschi-Bicchierai, who wrote about its penchant for periodically sending mystifying notifications that someone has been blocked from contacting you, even though nothing then prevents you from contacting that person directly. This has happened to me at least twice, and in both cases the person in question was already an approved contact, and said they had not attempted to reach out.

Although it would be nice if Apple’s Lockdown Mode sent fewer, less alarming and more informative alerts, the occasional baffling warning message is hardly enough to make me turn it off.