3124 stories
·
0 followers

'I leaked secrets': An AI agent confessed to its own data leak — but others got around corporate protections

1 Share
  • An AI assistant exposed company passwords across 10 sessions spanning three weeks
  • Another AI agent bypassed corporate restrictions using an alternative API connection
  • AI coding assistants can expose confidential credentials while completing routine software development tasks

Artificial intelligence agents are gaining access to company systems, internal documents, and sensitive credentials to complete tasks assigned by employees.

However, two incidents uncovered by NordLayer AI Scout reveal how ordinary instructions can expose confidential information or circumvent organizational security restrictions.

In one incident, an AI coding assistant even acknowledged that it had exposed sensitive credentials after repeatedly accessing company files.

AI assistant exposed company passwords

In this incident, a quality assurance engineer instructed an AI coding assistant to examine proposed software changes.

The project's instructions directed the assistant toward a GitLab access token stored inside an environment configuration file.

However, that file also contained production passwords, an authentication token for an internal human resources system, and credentials belonging to a business intelligence service.

The assistant repeatedly accessed the entire file across 10 separate sessions over three weeks, transmitting its contents to an external AI service each time.

The situation became worse when the assistant encountered a fresh project copy without the expected configuration file.

Rather than stopping its search, it retrieved the computer's entire collection of environment variables, exposing additional passwords and authentication credentials.

After recognizing what happened, the assistant acknowledged its mistake and recommended replacing the exposed credentials.

“I leaked secrets into this session's tool output… I'd rotate them,” the AI coding agent confessed in a session observed by NordLayer AI Scout.

Vakaris Noreika, a cybersecurity expert at NordLayer, said engineers frequently instruct AI assistants to retrieve credentials directly, creating unnecessary exposure.

He explained that assistants can repeat such behavior across subsequent tasks, allowing security problems to continue unnoticed.

AI agent bypassed company restrictions through an alternative connection

A separate incident involved an employee attempting to use an AI assistant to retrieve information from the company's Confluence documentation platform.

The organization's Model Context Protocol (MCP) connection appeared to prevent the assistant from accessing those internal pages.

Instead of treating the restriction as final, the employee asked whether creating an API token could provide another route into the documentation.

The assistant confirmed that this alternative approach worked, effectively circumventing the restriction imposed on its official connection.

The employee might have been seeking convenience rather than deliberately compromising company security.

However, using an alternative connection without authorization can undermine restrictions established by an organization's security team.

“Security controls only ever cover the cases someone foresaw. Every company has rules, but agents simply don’t know them, because they are not employees who were trained at the company,” says Noreika.

“This is a current technology limitation: Agents don’t retain rules between sessions, and sometimes even after getting specific instructions not to do something, they still do it.”

The incidents raise concerns about organizations granting automated assistants access to sensitive systems without adequate monitoring.

Organizations should separate sensitive credentials, store valuable passwords in secure vaults, track which resources AI agents can access, and monitor their activities.

Companies should also provide approved access routes rather than relying exclusively on restrictions that employees might circumvent.

Such measures could help security teams identify unusual access and prevent confidential information from reaching unauthorized destinations.

Google logo on a black background next to text reading 'Click to follow TechRadar'



Read the whole story
NerdsToGo
2 hours ago
reply
Share this story
Delete

As GTA VI arrives, McAfee warns of new malware-as-a-service targeting gamers and its own customers alike

1 Share
  • McAfee Labs details Marsalek, a free malware-as-a-service platform that hid infostealers in fake cheats for Roblox, Fortnite, Minecraft, CS2, and Grim Dawn
  • A hidden payload sent a second copy of stolen data to the operator, so the platform's own customers were robbed alongside their victims
  • The malware bypasses Chrome's App-Bound Encryption, raids 56+ crypto extensions and 12 desktop wallets, and backdoors Exodus and Atomic Walle

McAfee Labs has detailed Marsalek, a free malware-as-a-service (MaaS) platform kit which disguised password stealers as cheats for the likes of Roblox, Fortnite, and Counter-Strike 2 while quietly copying its customers’ stolen data to its own server.

Its tactic of targeting gamers is especially worrying following the recent release of GTA VI, the arrival of which capped months of fake “free game” downloads aimed at unsuspecting users looking to play it on PC.

McAfee Labs has now published a lengthy teardown of Marsalek, which let anyone sign up, build an infostealer, dress it up as a game cheat, and watch stolen passwords arrive on a dashboard.

A free-to-use malware product that makes victims of its users

Marsalek's pitch to aspiring criminals was that it cost nothing. McAfee says it only ever offered a free tier, with a premium tier perpetually promised, and that signing up did not even require a Discord or email account.

Customers got builders for PowerShell scripts, Python scripts, Windows executables, and ClickFix pages, the fake CAPTCHA trick that talks victims into pasting a malicious command into Windows themselves. A dashboard listed victims by IP address, plotted them on a global heatmap, and pushed alerts to Telegram, while an FAQ offered tips on using stolen Discord tokens and cookies, and on laundering crypto through a privacy-focused Monero node.

The catch is what McAfee calls a dual-hook design: Alongside its own stealer, Marsalek deploys a fourth-stage payload called PureLog, which ships a second copy of everything it harvests to a separate server. In practice, the platform's operator could take whatever customers stole, without those customers ever knowing.

McAfee adds that paying for the eventual premium tier would not have helped, because the owner could pull the plug on the command-and-control server at any time and walk off with both the credentials and the subscription fees alike.

PureLog, more widely tracked as PureLogs, is not bespoke. It belongs to a commercial malware family sold by a developer known as PureCoder, which Check Point has documented alongside the PureCrypter obfuscator and PureRAT. McAfee puts PureLog subscriptions at as little as $59 per month.

The free service, in other words, appears to have been underwritten by someone paying for an off-the-shelf stealer to skim victims gathered by other people's campaigns. The old line about free products being paid for with the user applies with unusual force here: the would-be hackers were as much the product as their targets.

There is a small irony in the sales pitch, too: Marsalek's FAQ promised it could raid every Gecko-based browser, but McAfee found it only targets Firefox. Even criminal software, it seems, overstates its spec sheet.

When it works as intended, though, Marsalek is thorough. McAfee says it targets more than 32 Chromium-based browsers and sidesteps App-Bound Encryption in Chrome, Edge, and Brave by injecting a DLL into a suspended browser process, a protection that infostealer developers claimed to have beaten not long after Google introduced it.

It also decrypts saved Firefox logins, checks 26 locations for Discord tokens and validates them against Discord's own API, and grabs Telegram sessions and Mullvad VPN account metadata.

Crypto holders get special attention. The malware hunts through more than 56 browser extensions and 12 desktop wallets, and backdoors the Exodus and Atomic Wallet desktop apps by swapping out their .asar archives. It escalates privileges with a CMSTP-based UAC bypass, adds more than 23 Windows Defender exclusions, and persists through the Startup folder and scheduled tasks.

PureLog then layers on a screenshot, webcam details, and hardware IDs, relaunching every five minutes via a scheduled task named RunetimeBroker.exe, one stray letter away from the genuine Windows RuntimeBroker process.

McAfee's telemetry shows infections concentrated in the US, followed by Germany, the UK, India and Brazil. It does not publish a victim count, so its published 1,560 figure describes would-be attackers rather than people infected. Its broader conclusion is blunt and starker, however: "Infostealer logs have become the backbone of cybercrime in 2026."

GTA VI is here, and the scam economy came with it

McAfee had already issued its own warning about GTA VI scams, flagging offers to play before launch, VIP early access, secret beta downloads, discounted preorder keys, and crypto-only payments.

"Every major game launch brings a noticeable increase in gaming-themed scams," said McAfee threat researcher Charles McFarland.

For context, three days before release, NordVPN's threat intelligence team found a $200 scam kit on the dark web, payable only in Bitcoin or Tether and built entirely around a fake PC release. Other finds followed the same pattern.

Cynet traced a fake GTA VI download from a Google Colab notebook to a bogus "Windows Software Hub" page, where a supposed 72MB installer turned out to be a 796KB executable carrying the Warden infostealer. A bundled README told victims to run it as administrator and add an antivirus exclusion, the same Defender-dodging step Marsalek takes automatically.

Netcraft counted 121 newly registered domains using the name of CyberLeek, the persona behind this year's gameplay leaks, including one offering a "GTAVI.exe" that 34 of 68 VirusTotal engines flagged.

Furthermore, on launch day, Malwarebytes tracked lures ranging from a free-copy giveaway that harvests Google passwords through a cloned sign-in page to a GTA VI-branded crypto site whose "buy" button opens a wallet-connection request.

Earlier waves were no gentler. Malwarebytes researcher Stefan Dasic called the game "the perfect bait" when scam sites began selling fake early access for cryptocurrency, while Huntress found fake installers padded beyond 100GB carrying the NjRAT and DCRat remote access trojans, the Mercurial Grabber stealer, and Chaos ransomware.

The most useful fact for players is still a simple one: GTA VI runs only on PS5 and Xbox Series X|S, and ESET points out that Rockstar has not announced a PC version, demo, beta, or early-access program. Even Xbox Cloud Gaming streams the game only to Xbox consoles at launch, not to PCs.

The longer-term concern is what follows a real PC launch. Marsalek's lures were cheats, executors, and spoofers for games with active modding and competitive scenes, and nothing McAfee describes ties its builders to any single title.

Its own analysis of filenames and download pages implies that the game-specific part was the costume, not the code. The NordVPN kit shows the GTA VI costume is already on sale; a PC version, and the modding scene sure to follow, would supply the cheats and mods that Marsalek and WeedHack fed on.

McAfee's advice is familiar but worth repeating. Treat free cracks, cheats, and executables shared on Discord or GitHub with suspicion, and never disable antivirus or add an exclusion to run a tool it flags. If you suspect an infection, a dedicated malware removal tool is a starting point, but stolen session cookies and wallet credentials outlive the malware itself.

McAfee, which sells consumer security software, unsurprisingly recommends its own products, though a reputable antivirus suite from any vendor is the baseline. For anyone tempted by Marsalek's side of the transaction, the research carries a warning of its own: the platform was built to rob its customers too.



Read the whole story
NerdsToGo
2 hours ago
reply
Share this story
Delete

FBI reportedly hacked again, this time targeting nationwide police database

1 Share
FBI systems have reportedly been breached for the second time in two months, this time targeting the agency's National Crime Information Center (NCIC), which is used to share information between local, state, and federal law enforcement agencies. [...]
Read the whole story
NerdsToGo
2 hours ago
reply
Share this story
Delete

Microsoft Entra Connect Health flaw exploited to steal tenant data

1 Share
Microsoft is warning organizations that it suffered a data breach after attackers exploited a vulnerability in its Entra Connect Health service to gain unauthorized access to administrator email addresses and tenant configuration information. [...]
Read the whole story
NerdsToGo
2 hours ago
reply
Share this story
Delete

Kidnapping scams on the rise in Tennessee, TBI warns - WKRN News 2

1 Share
Kidnapping scams on the rise in Tennessee, TBI warns  WKRN News 2
Read the whole story
NerdsToGo
2 hours ago
reply
Share this story
Delete

Too Lost - 724,323 breached accounts

1 Share
In November 2026, an unauthorised party claimed to have obtained data from the music distributor for independent artists, Too Lost. The data contained 724k unique email addresses belonging to artists, their collaborators and fans. Some records also contained bcrypt password hashes, while a small subset included partial credit card data (last 4 digits and card type), identity verification information (in some cases, passport, driver's licence or national ID numbers), and payouts made to artists. Too Lost advised that its investigation remains ongoing, external forensic specialists have been engaged, and measures have been taken to secure affected systems and credentials. The company is also working with advisors to coordinate communications and support for affected individuals.
Read the whole story
NerdsToGo
1 day ago
reply
Share this story
Delete
Next Page of Stories