New research reveals that AI safety refusal lives in a thin neural layer, highlighting the critical need for external, multi-layered security. The post Perturbation Probing: A New Diagnostic for the Fragility of LLM Safety appeared first on Unit 42.
Love Electric’s alleged data breach exposes sensitive driver data and highlights the identity risks created by third-party salary sacrifice providers. A seller on an English-language data-breach forum claimed on August 26 that they had obtained the driver database of Love Electric, a UK broker that runs electric-vehicle salary sacrifice schemes. The seller, operating under the name “seraphims”, offered 877,000 records for $600 in cryptocurrency, with the price negotiable. That headline number needs a qualification. Ransomnews researchers examined a 999-row sample published with the listing and found strong evidence that the sample came from a genuine production database, but the claimed 877,000 records remain unverified. Love Electric had been contacted for comment at the time of publication. “The 999 rows split into 731 primary drivers and 268 additional named drivers. There are exactly 731 distinct quote IDs, and every quote has exactly one primary driver, 731 out of 731. All 268 additional drivers reference a quote that exists in the file, with no orphans.” reads the Ransomnews’s report. “The consent flag governing National Insurance processing is empty for precisely the 268 additional drivers and populated for every one of the 731 primary drivers. National Insurance numbers appear on primary drivers only. Those are constraints enforced by whatever application wrote the data, and they hold across all 999 rows without exception.” That matters because fabricated breach databases often fall apart when researchers examine the relationships between records. Here, the relationships hold: additional drivers point to existing quotes, the National Insurance processing flag appears where the underlying application logic would expect it, and National Insurance numbers appear only on primary drivers. Love Electric Financial Services Limited is an Edinburgh-based company that provides EV salary sacrifice administration, credit broking and related services to employers across the UK. Companies House identifies it as an active Scottish company, registered under number SC374952, while Love Electric’s own documentation confirms that it operates as a salary sacrifice administrator and FCA-regulated credit broker. The business model explains why its systems can contain information that many people would normally associate with payroll departments. Salary sacrifice arrangements require the broker to identify employees, process the scheme and support insurance and tax-related requirements, so the underlying records can include National Insurance numbers and driving licence information. Love Electric’s privacy policy also says it processes personal information under UK data protection law. “The seller published a free sample as a 999-row CSV with 24 columns, named in a way that indicates an export of a dbo.drivers table from a Microsoft SQL Server database.” continues the report. “The columns are: id, quote_id, user_id, title, first_name, last_name, email, phone_number, date_of_birth, address, address2, city, country, postcode, national_insurance_number, driving_licence_number, driving_licence_country, allow_processing_national_insurance, primary, created_at, updated_at, deleted_at, weekly_hours and occupation_id.” The fields included names, email addresses, phone numbers, dates of birth, addresses, postcodes, National Insurance numbers, driving licence numbers, quote IDs and consent-related fields. The data wasn’t uniformly populated, which is another reason the sample looks like a real production export rather than a synthetic dataset. Around 71% of the rows lacked a name, address or city, 74% had no phone number, and 85% had no National Insurance number, while 147 records contained a National Insurance number and 287 contained a driving licence number. The geography also made sense. Postcodes clustered around Edinburgh and central Scotland, with records extending into England, while dates of birth ranged from 1946 to 1999 and concentrated around people born in the 1970s and 1980s. Ransomnews then tested the driving licence numbers against the structure used by UK licences. Of the 108 full-length licence numbers in the sample, 98.1% had a surname block matching the surname in the corresponding record, 97.2% had an initial matching the first name, and 78.7% contained a date-of-birth encoding that matched the stored date of birth. The imperfections were just as interesting. Licence numbers appeared in many different lengths, only 53% of the National Insurance numbers present matched the expected HMRC format, and phone numbers appeared in three different formats. That kind of mess is exactly what you’d expect from users entering information into a live form, abandoning fields or making mistakes. “A UK driving licence number is not random. Characters one to five are the holder’s surname padded with the digit 9, characters six to eleven encode the date of birth with 50 added to the month for female holders, and character twelve is the first-name initial. Of the 108 full-length licence numbers in the sample, 98.1% have a surname block matching the record’s own surname field and 97.2% have an initial matching the first-name field. 78.7% carry an encoded date of birth matching the date_of_birth column.” state the researchers. “The roughly one in five that disagree are consistent with people mistyping a sixteen-character string into a web form, which is exactly the error rate real self-reported data carries and synthetic data does not.” The email addresses provided another useful check. Most belonged to corporate domains associated with identifiable UK employers, Love Electric itself or a European software consultancy, rather than being dominated by Gmail and other consumer providers. That fits the way workplace salary sacrifice schemes operate because employees generally enter through their employer. The database structure also looked normal. It included a soft-delete field, integer foreign keys for quotes and occupations, regional values for Scotland, England, Wales and Northern Ireland, and even a Jane Doe test record that appears to have survived from the system’s setup. Real databases leave fingerprints like these. Synthetic ones tend to be suspiciously tidy. The seller’s own history is less convincing than the database evidence. The “seraphims” account was created on July 22, 2026 and had published nine data listings by August 26, roughly two per week, but several of those listings were described by the seller as scrapes rather than breaches. The Love Electric listing had attracted no replies and only 52 views when Ransomnews checked it. The account’s reputation score of 30 came from a single positive vote posted on August 17 with the comment “keep scrapping”, which says more about the seller’s scraping activity than about a track record of selling verified breach data. That doesn’t make the Love Electric sample fake. The technical evidence points in the opposite direction. It does mean we should separate two questions: whether the sample appears genuine, and whether the seller really controls 877,000 records. The second claim is much harder to establish. Ransomnews saw only 0.11% of the alleged dataset, and the sample shows that database rows don’t map directly to individual people. The 999 rows corresponded to 731 quotes and only 58 distinct surname-and-date-of-birth combinations, with one person appearing 48 times. There’s another clue in the filename. The db2_ prefix suggests that the table may belong to one database among several, so the seller’s 877,000 figure could refer to a larger collection rather than this particular driver table. At this point, treating 877,000 as a confirmed number would go beyond the evidence. The timeline also deserves caution. The seller claimed an August 2026 attack involving a zero-day in a third-party system, but every sample row carries a created_at timestamp within a six-second window on August 14, 2022. That pattern looks like a bulk migration into a newer platform rather than 999 people signing up within six seconds. It tells us that the database contains legacy records. It doesn’t tell us when someone took them, how they obtained them or whether the seller’s claimed zero-day played any role. That last point matters for incident response. “Third-party zero-day” is a claim made by the seller, not an established fact, and attackers have plenty of reasons to make an access story sound more sophisticated than it was. Ransomnews says it has not verified that route. For the people potentially affected, the more immediate issue is the nature of the exposed information. A password can be replaced. A National Insurance number and a driving licence number can’t simply be rotated because someone has posted them for sale. That creates an obvious phishing risk. Someone who has a person’s name, date of birth, address, employer and National Insurance number can construct a much more convincing message pretending to be HMRC, a payroll team, an employer or a vehicle leasing provider. Drivers who used Love Electric should therefore treat unexpected messages about their vehicle scheme, payroll or tax affairs with suspicion. They should verify the sender through a phone number or website they already trust rather than using contact details contained in the message. Love Electric’s current public material says it works with more than 1,500 companies, which gives some sense of the scale of the service’s employer-facing footprint. That makes supplier security worth treating as an operational issue rather than a procurement checkbox. If a third-party provider holds National Insurance numbers, driving licence numbers, employment information and contact data, its security controls deserve the same level of scrutiny as any other system handling high-value identity information. The seller’s alleged third-party entry point is therefore worth investigating, but it shouldn’t become the centre of the story before someone verifies it. The useful evidence is already sitting in the sample: a production-style schema, consistent relationships, realistic user errors and identifiers that line up with real-world records. The other important detail is the price. $600 for a database containing National Insurance and driving licence numbers isn’t a serious valuation of the data. It’s a price designed to make a quick sale. And if the seller really has more than the published sample, multiple buyers could obtain the same information. That’s where the risk gets worse: the exposure doesn’t end with the first person who downloads it. Ransomnews says it did not access or probe Love Electric’s systems, purchase the full dataset, or validate the exposed identifiers against live services. It analysed the free sample, removed identifying information from its publication and notified Love Electric before publication, offering to share the listing and sample with the company’s incident-response team. Follow me on Twitter: @securityaffairs and Facebook and Mastodon Pierluigi Paganini (SecurityAffairs – hacking, data breach)
PaperCut has released a second emergency security update for two actively exploited vulnerabilities in its PaperCut NG and MF print management software after researchers discovered multiple ways to bypass the initial fixes. As BleepingComputer reported yesterday, PaperCut warned that hackers were exploiting a vulnerability in zero-day attacks against customer servers and released an initial emergency patch for PaperCut NG/MF versions 25 and 26. At the time, however, the company had not disclosed CVE identifiers or technical details about the vulnerabilities, saying it was withholding information while it investigated the attacks and gave customers time to apply emergency fixes. PaperCut has now shared technical details and CVE identifiers for the two vulnerabilities, tracked as CVE-2026-82078 and CVE-2026-81578. These vulnerabilities can be chained to bypass authentication and execute code on vulnerable servers.
A maximum-severity vulnerability in the GiveWP plugin for WordPress allows an unauthenticated attacker to execute arbitrary commands on the hosting server. The security issue is identified as CVE-2026-82222 and affects GiveWP through version 4.16.7.1. It was reported by bug researcher Udin Chan on July 28 through the Patchstack vulnerability intelligence platform. The GiveWP plugin has more than 100,000 installs and allows collecting donations and managing fundraising campaigns. Patchstack researchers explain that exploiting the vulnerability is possible by chaining three distinct issues: Successful exploitation depends on the attacker having an account on the target site. However, Patchstack says that an exposed unauthenticated registration action allows creating an account even if registration is disabled.
A 68-year-old has been sentenced in the U.K. to more than six years in prison for operating an illegal IPTV (Internet Protocol Television) service that generated £980,812 ($1.3 million) over three years. An investigation by the Police Intellectual Property Crime Unit (PIPCU) at the City of London Police found that Milan Ibrahim ran a "sophisticated operation" that provided illegal IPTV services to users in the UK and abroad. According to PIPCU, Ibrahim sold illegal broadcasts from major rights holders such as the BBC, ITV, Sky, the Premier League and the Motion Picture Association. The police seized and shut down all servers they found during the operation, disrupting the illegal streams that users of the IPTV service received. “The investigation revealed that the business operated on 80 servers from premises in Chorley and generated £980,812 over a three-year period,” announced the City of London Police.
AI is accelerating vulnerability discovery, putting pressure on systems built to enrich, prioritize, and remediate flaws at a slower pace. Action1 explains why defenders increasingly need to correlate multiple intelligence sources and turn vulnerability data into faster remediation.
U.S. Cybersecurity and Infrastructure Security Agency (CISA) adds ownCloud, Linux Kernel, and JFrog Artifactory flaws to its Known Exploited Vulnerabilities catalog. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added the following vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog: CVE-2023-49105 ownCloud Improper Authentication Vulnerability CVE-2026-53362 Linux Kernel Unspecified Vulnerability CVE-2026-66384 JFrog Artifactory Improper Limitation of a Pathname to a Restricted Directory Vulnerability CVE-2023-49105 (CVSS score of 9.8) is an improper-authentication flaw in ownCloud Server’s WebDAV functionality. An unauthenticated attacker who knows a victim’s username can read, alter, or delete that user’s files when no signing key is configured, because the application can accept pre-signed URLs without properly validating that condition. The issue affects ownCloud core versions 10.6.0 through 10.13.0, before version 10.13.1. Organisations running affected ownCloud instances should treat remediation as urgent, review exposed WebDAV services, and investigate unusual file access, deletion, modification, or downloads. CVE-2026-53362 (CVSS score of 7.8) is an out-of-bounds memory-write vulnerability in the Linux kernel’s IPv6 networking subsystem. A local attacker who can create UDP sockets could exploit an incorrect parameter-length calculation during fragmented IPv6 packet handling to overwrite kernel memory, potentially causing a system crash, corrupting data, or escalating privileges.access. Shortly after the Hugging Face incident became public, OpenAI said its models had identified and exploited a zero-day vulnerability in JFrog’s Artifactory package registry manager. The report also describes a separate incident on July 19. In that case, some AI agents exploited the Linux kernel vulnerability, CVE-2026-53362, to gain higher privileges inside an OpenAI environment. The agents detected that the Linux kernel on their machine was vulnerable, found a public exploit, modified it to work in their environment, and used it to gain root access to the underlying worker node. This allowed them to escape the Artifactory container and move laterally to other parts of the connected environment. CVE-2026-66384 (CVSS score of 5.3) is a path-traversal vulnerability in JFrog Artifactory’s Docker cache handling. Under specific remote-repository conditions, an authenticated user can manipulate a file path and write data outside the intended cache directory, potentially affecting other locations on the Artifactory host. According to Binding Operational Directive (BOD) 22-01: Reducing the Significant Risk of Known Exploited Vulnerabilities, FCEB agencies have to address the identified vulnerabilities by the due date to protect their networks against attacks exploiting the flaws in the catalog. Experts also recommend that private organizations review the Catalog and address the vulnerabilities in their infrastructure. CISA orders federal agencies to fix the flaw CVE-2026-66384 by September 10, while the remaining must be addressed by August, 30, 2026. Follow me on Twitter: @securityaffairs and Facebook and Mastodon Pierluigi Paganini (SecurityAffairs – hacking, CISA)
Over 8,300 Internet-exposed Gitea instances are still unpatched against a critical security flaw exploited in ongoing remote code execution attacks, according to cybersecurity watchdog Shadowserver. The code injection vulnerability (CVE-2026-60004) targeted in these attacks was reported by Salesforce security researcher Shai Rod, and it allows authenticated attackers to execute arbitrary shell commands with the privileges of the Gitea service account by submitting malicious patches via the diffpatch API endpoint. While successful exploitation requires repository write access to repositories hosted on vulnerable servers, Gitea comes with self-registration enabled by default, allowing unauthenticated attackers to register an account, create a new repository, and trigger the vulnerability without prior credentials.
Hasbro, one of the world's largest toy and game companies, has disclosed that attackers have accessed the personal and financial information of an undisclosed number of employees. Founded in 1923, Hasbro is a publicly traded American multinational entertainment conglomerate on the NASDAQ and owns many brands, including Monopoly, Clue, Nerf, Transformers, Play-Doh, Peppa Pig, Scrabble, Magic: The Gathering, Dungeons & Dragons, and many others. The company has filed data breach notification letters with the Massachusetts Attorney General's Office, but didn't disclose the total number of affected individuals or when the incident was detected. "The information involved varied by individual but may have included your name and one or more additional personal information elements such as email, address, phone number, national ID number, or financial information," Hasbro said.
BlueDelta (APT28) uses webhook.site and Microsoft Edge to hide HOOKEDGE espionage traffic targeting European governments. Recorded Future’s Insikt Group documented a campaign by BlueDelta, the Russian GRU-linked group that overlaps with the group APT28, running an entire espionage operation against European government targets using webhook.site, a service built for developers to test HTTP requests, as its command-and-control backbone. The campaign ran from late September 2025 through early April 2026, targeting government and diplomatic organizations in Romania, Spain, and Türkiye. “The campaigns delivered a lightweight Windows batch-script backdoor, dubbed “HOOKEDGE,” via macro-enabled Microsoft Word documents using diplomatic-themed lures, including material impersonating Spain’s Ministry of the Presidency, Justice and Relations with the Cortes, created shortly after a September 2025 meeting between Spanish and Moldovan officials.” states the report by Insikt Group. “Insikt Group assesses with moderate confidence that this activity was conducted by BlueDelta (which overlaps with APT28, Fancy Bear, and Forest Blizzard), a Russian state-sponsored threat group attributed to the Main Directorate of the General Staff of the Armed Forces of the Russian Federation (GRU).” The attackers used a fairly old-school trick: macro-enabled Word documents. One of them looked like a real meeting agenda from Spain’s Ministry of the Presidency and appeared shortly after an actual meeting between Spanish and Moldovan officials. The timing was interesting because it came just before Moldova’s September 2025 parliamentary elections. The document looked credible enough to make the phishing attempt believable. The backdoor, called HOOKEDGE, is surprisingly simple. It uses a Windows batch script and relies on two webhooks. One receives commands, while the other sends the stolen data back to the attackers. Every 30 minutes, a scheduled task downloads a command file through Microsoft Edge, runs it and sends the results to another endpoint. The malware uses Edge to make the traffic look like normal web activity, making it harder for security tools to spot. That reliance on a real browser for both tasking and exfiltration is the cleverest part of the whole design. “A notable aspect of HOOKEDGE is its use of msedge.exe for both tasking and exfiltration. By generating network traffic through a legitimate web browser rather than a commonly abused LotL binary (LOLBin) or a custom binary, the malware blends its communications with normal enterprise browsing activity.” continues the report. Recorded Future explains, and that’s really the entire evasion strategy in one sentence: don’t build something exotic, just make your traffic look exactly like an employee checking a website. BlueDelta didn’t build HOOKEDGE from scratch either. It shares deep code and structural overlap with HEADLACE, a backdoor the same group used years earlier, right down to identical JavaScript variable names and the same base64 encoding scheme for automated downloads. “HOOKEDGE’s code and structural design have significant overlap with HEADLACE, a backdoor used by BlueDelta in previous campaigns.” states the report. Recorded Future assesses with moderate confidence that HOOKEDGE is a direct evolutionary successor, maintained by the same operators rather than a fresh tool built by a different team, which fits BlueDelta’s long-documented habit of refining working tradecraft instead of reinventing it. The operation also included a triage mechanism worth understanding on its own. Once a victim showed signs of being worth deeper attention, active communication with the initial webhook endpoints, BlueDelta deployed a second HOOKEDGE payload configured to check in every five minutes instead of thirty, giving operators much faster interactive control over higher-value targets. That two-tier setup also solved a practical infrastructure problem: webhook.site’s free tier caps out at 100 requests per endpoint, so spreading routine and high-priority tasking across separate endpoints kept any single one from getting exhausted mid-operation. BlueDelta kept tuning the operation continuously rather than treating it as a finished product. Beaconing intervals stretched from 30 minutes to 61, deliberately timed to slip past sandbox environments that typically only watch a sample’s behavior for an hour. The group added canary tracking pixels named things like mailopened.jpg and docopened.jpg to monitor exactly when a phishing email got opened versus when the document itself got opened versus when macros actually executed, essentially building analytics for their own phishing funnel. “The malicious document also contains a hidden image referencing a remote webhook URL: hxxp://webhook[.]site/62114596-33f5-47fb-9012-0223529e5a13/docopened[.]jpg. This serves as a document-open “canary,” alerting BlueDelta operators when a victim opens the lure. Later variants used the filename doc.jpg in place of docopened.jpg.” states Insik. “Insikt Group also identified webhooks using the filename mailopened.jpg, indicating that BlueDelta likely used a similar canary mechanism to monitor when recipients opened phishing emails, providing operators with visibility into campaign delivery success before any payload execution.” That kind of detailed tracking of what victims do would look very familiar to a marketing team. The difference is that here the targets are victims, not customers. For defenders, the useful indicators are quite clear. Block macros in documents downloaded from the internet, monitor scheduled tasks that launch scripts from user-writable folders, and flag Microsoft Edge running in headless mode or making automated connections to file-sharing and webhook services that the organization does not normally use. You don’t need sophisticated tools to spot these behaviors. Even a well-configured monitoring system should be able to detect them early. The worrying part is that a persistent, state-backed group can still rely on a small set of behaviors that defenders can identify and block. For defenders, the actionable pieces here are concrete rather than abstract. Block macro execution from documents that arrived over the internet, watch for scheduled tasks spawning script interpreters from user-writable folders, and specifically flag Microsoft Edge running in headless mode or making automated requests to file-hosting and webhook services your organization doesn’t actually use for anything legitimate. None of this requires exotic tooling to catch, which is oddly reassuring: a threat group this persistent and state-backed is still, at its core, relying on the same handful of detectable behaviors that a properly configured monitoring setup would catch on day one. “BlueDelta is likely to continue conducting initial access campaigns against European government and diplomatic organizations in support of Russian intelligence collection. Given the enduring strategic importance of European governance, NATO-related affairs, and diplomatic engagement with former Soviet republics, the intelligence requirements driving this activity are unlikely to diminish in the near term.” concludes the report. Follow me on Twitter: @securityaffairs and Facebook and Mastodon Pierluigi Paganini (SecurityAffairs – hacking, BlueDelta)
Confirm this action.
Leaving now will discard your changes.