World

Anthropic has disclosed that it identified and disrupted what it describes as industrial-scale illicit distillation attacks targeting its Claude model. According to the company, the activity originated from seven labs based in China, among them Alibaba, Moonshot, DeepSeek, Z.ai (also known as Zhipu), and MiniMax.

It is worth noting that knowledge distillation is not inherently malicious. It is a legitimate machine learning training technique in which a large, capable AI model acts as a teacher to transfer knowledge to another model. The concern in this case stems from how the technique was allegedly applied and at what scale, which Anthropic characterizes as illicit.

Key details as reported:

• Seven China-based labs were named in connection with the campaign.
• Named entities include Alibaba, Moonshot, DeepSeek, Z.ai (aka Zhipu), and MiniMax.
• Anthropic states it identified and disrupted the attacks against Claude.
• The activity is described as industrial-scale distillation.
• Knowledge distillation itself remains a legitimate training method; the dispute concerns its unauthorized use in this context.

The report does not include specific CVE identifiers, advisory numbers, or indicators of compromise, so none are listed here. Organizations evaluating their own exposure to similar model-abuse campaigns should monitor vendor advisories and terms-of-service enforcement actions rather than rely on signature-based detection alone.

Source: The Hacker News

Do you think API-level rate limiting and output watermarking are enough to deter distillation attempts, or is stronger contractual and technical enforcement needed?

Cybercriminals are increasingly using AI to close the trade-off between the scale of an email campaign and its believability. According to Dark Reading, one threat actor managed to generate 1 million personalized fraud emails in just 3 days, a volume that would previously have required sacrificing the tailored, credible feel that makes phishing convincing.

The significance here is the shift in the economics of email-based attacks. Personalization has traditionally been the bottleneck for large campaigns, since crafting believable messages for each target takes time and effort. AI-assisted tooling removes that constraint, allowing attackers to push out high volumes of individually tailored messages without the usual drop in quality. The report frames this as the end of having to choose between volume and credibility in malicious email operations.

Practical takeaways for defenders:

• Treat personalization in messages as weak evidence of legitimacy, since AI-generated content can be tailored at scale.
• Reinforce user awareness that convincing, contextually relevant emails are not inherently trustworthy.
• Prioritize detection and filtering controls that do not depend on spotting generic, poorly written content.

Source: Dark Reading

Has your organization adjusted its email security controls in response to AI-generated phishing at this scale?

GitLab has shipped patches for multiple security issues, including a maximum-severity flaw that attackers began probing in the wild within hours of public disclosure.

The most serious issue is CVE-2026-85706, which carries a CVSS score of 10.0. It is a path traversal vulnerability in the repository commits API that could let an unauthenticated user read arbitrary files from the GitLab server.

Details on affected versions and official remediation guidance were not included in the source report, so administrators should consult GitLab's own advisory and apply the available patches as soon as possible. Given the maximum severity rating and evidence of active probing, treating this as an urgent patch is advisable.

Source: The Hacker News

Has anyone seen probe attempts against their GitLab instance yet, and how quickly are you planning to roll out the fix?

The Florida Department of Highway Safety and Motor Vehicles (FLHSMV) has confirmed a data breach affecting its DAVID driver database. According to the agency, attackers gained access to the system using credentials belonging to a police department employee.

The breach was carried out through a stolen account tied to law enforcement, which gave the attackers access to the driver records stored in the DAVID system. FLHSMV has acknowledged the incident and confirmed that the unauthorized access occurred via these compromised credentials.

Details on the full scope of the breach, including how many records were affected or when the access took place, have not been disclosed in the available reporting.

Key facts:

• Affected system: DAVID driver database
• Affected organization: Florida Department of Highway Safety and Motor Vehicles (FLHSMV)
• Attack method: access via stolen credentials belonging to a police department employee

This incident highlights the ongoing risk posed by credential theft, particularly when accounts with access to sensitive government systems are targeted.

Source: BleepingComputer

Does your organization enforce phishing-resistant MFA for accounts with access to sensitive databases like this one?

Anthropic has disclosed that multiple threat groups attempted to abuse its Claude AI model for malicious purposes, including financially motivated actors and state-sponsored espionage groups linked to Russia and China.

One notable campaign involved using Claude to extract secrets from 1.8 million Android apps. This highlights how large language models are increasingly being leveraged by attackers to accelerate and scale up their operations.

The report underscores a growing trend of AI-assisted cyberattacks, where threat actors exploit AI capabilities for tasks such as vulnerability discovery, credential harvesting, and data extraction at scale.

• Threat groups involved include financially motivated actors and state-sponsored espionage groups
• Nations linked to the activity include Russia and China
• 1.8 million Android apps were targeted for secret extraction
• The AI model abused was Claude

Source: Unknown

How is your organization assessing the risk of AI-assisted attack techniques in your threat model?

Microsoft is warning that threat actors tied to extortion groups including ShinyHunters and Helix are running social engineering campaigns built around passkey and single sign-on themes to break into corporate Microsoft accounts and exfiltrate data from Microsoft 365 services.

The attacks rely on convincing users to hand over access under the guise of passkey enrollment or SSO-related activity, rather than exploiting a software flaw. That means the entry point is the user, and the payoff for the attackers is access to cloud-hosted corporate data.

Key points to keep in mind:

• The campaigns are attributed to multiple extortion gangs, among them ShinyHunters and Helix.
• Targets are corporate Microsoft accounts, with data theft focused on Microsoft 365 services.
• The lures center on passkey and single sign-on themes, which makes them harder to spot since both are legitimate, everyday authentication topics.
• There is no indication in the reporting of a specific CVE or product vulnerability being exploited here; this is a social engineering problem, not a patchable bug.

Because these attacks abuse trust in authentication features rather than technical weaknesses, the usual defenses matter more than ever: strong user awareness around credential and passkey requests, conditional access and phishing-resistant authentication policies, and monitoring for unusual sign-ins or data access in Microsoft 365. Organizations should also review how passkey rollouts are communicated internally, since attackers are mimicking exactly that kind of messaging.

Source: BleepingComputer

Has your organization seen any passkey or SSO-themed phishing attempts lately, and how are you handling them?

Threat actors are chaining together multiple JFrog Artifactory vulnerabilities to compromise self-hosted servers, bypass authentication, escalate to administrative privileges, and ultimately deploy a Rust backdoor on victim systems, according to new research.

The attacks target self-hosted Artifactory instances that remain unpatched against known critical and high-severity flaws. By combining these issues, attackers can move from initial unauthenticated access all the way to full administrative control, which they then leverage to plant persistent malware.

Key points from the report:

• The campaign exploits both critical and high-severity flaws in JFrog Artifactory
• Attackers bypass authentication as part of the chain
• They escalate privileges to gain administrative access
• A Rust backdoor is deployed on successfully compromised self-hosted servers
• Only self-hosted instances are affected — the flaws are not present in cloud-hosted deployments

If your organization runs a self-hosted JFrog Artifactory instance, review your current patch level against the vendor's latest advisories and confirm that the instance is not reachable from the public internet unless absolutely necessary.

Source: BleepingComputer

Has anyone here audited their self-hosted Artifactory deployment since this campaign came to light, and what steps are you taking to lock it down?

Threat actors are increasingly abusing trusted AI platforms as an attack surface, leveraging the reputation of services like Claude to distribute malware and trick users, according to an analysis by Huntress.

The campaigns observed by Huntress target AI users through several distinct vectors:

• Weaponized Claude Artifacts — malicious content hosted within the platform's artifact feature
• Shared AI conversations — poisoned or crafted chat threads used as lures
• Sponsored search results — paid ads directing victims toward malicious destinations
• ClickFix-style lures — fake error or verification prompts that trick users into running malicious commands themselves

The common thread across these techniques is abuse of user trust: because the content appears to originate from or relate to legitimate AI services, victims are more likely to interact with it without suspicion. Search results are also being poisoned to push malicious content toward users searching for AI-related tools and information.

Anyone using AI platforms in their workflow should treat shared artifacts, conversation links, and AI-themed search results with the same caution they would apply to any unsolicited file or link.

Source: Unknown

Has your organization implemented any controls around how employees interact with AI platforms and shared AI-generated content?

GitLab is warning administrators to patch their servers immediately following the disclosure of a maximum-severity path traversal vulnerability tracked as CVE-2026-85706.

Path traversal flaws of this kind can allow an attacker to reach files and directories outside the intended scope, which is why GitLab is treating this one as urgent. Given the maximum severity rating, administrators should prioritize patching over routine maintenance windows.

Details on affected versions, exploitation status, and any indicators of compromise were not included in the available reporting, so check GitLab's official security advisory for the authoritative version list and fixed releases before upgrading.

• Apply the available GitLab security update as soon as possible.
• Verify your deployed version against the official advisory rather than relying on secondary coverage.
• Monitor GitLab's channels for any follow-up guidance or revised severity information.

Source: BleepingComputer

Has your organization already applied this update, or are you still waiting on a maintenance window?

A Ukrainian national has been sentenced to four years in prison for his involvement in Conti ransomware operations conducted between 2021 and 2022. The sentence closes out a case tied to one of the more prolific ransomware groups of that period, which targeted organizations across multiple sectors and regions before its infrastructure was disrupted and its members scattered.

Conti operated as a ransomware-as-a-service operation, with affiliates carrying out intrusions while core members maintained the malware, negotiation portals, and leak site. The group was known for double extortion tactics, stealing data before encrypting systems and threatening to publish it if victims did not pay.

Details on the specific charges, the defendant's role within the group, and the jurisdiction where the sentence was handed down were not elaborated in the source report beyond the prison term and the timeframe of the attacks.

Key points:

• Sentence: four years in prison
• Group involved: Conti ransomware
• Attack period: 2021 to 2022
• Defendant: Ukrainian national

Source: BleepingComputer

Does your organization treat Conti-era indicators as still relevant in current threat hunting, or have you retired them?

Trezor has disclosed that a phishing campaign targeting its customers earlier this week reached 347,000 email addresses and affected 2,500 users who clicked an embedded malicious link.

The attack followed a breach at Brevo, a third-party email service provider, which enabled the threat actors to send fraudulent emails impersonating the hardware wallet vendor.

• 347,000 email addresses were targeted in the phishing campaign
• 2,500 users clicked the malicious link embedded in the emails
• The breach originated from Brevo, a third-party email service provider

Source: BleepingComputer
Were you or your organization impacted by this campaign, and are you reviewing third-party email providers in your supply chain?

Check Point has patched two critical vulnerabilities in how its firewall and management products handle VPN certificates. Both flaws carry a CVSS score of 9.8, and according to the company, they could allow an unauthenticated remote attacker to execute code — though only "under specific conditions" that Check Point has not disclosed.

The two issues are distinct in scope:

• One flaw affects Check Point Security Gateways, the company's firewall appliances.
• The second affects those same gateways as well as the Security Management side of the product line.

The article does not provide specific CVE identifiers for either vulnerability, so no advisory numbers are cited here. Administrators should refer to Check Point's official guidance for exact patch details and affected builds.

Because exploitation requires conditions Check Point has chosen not to describe, the practical risk is difficult to assess from the public disclosure alone. Even so, a 9.8 rating and unauthenticated RCE potential make prompt patching the sensible default for anyone running the affected gateway or management components.

Mitigation steps:

• Apply the vendor's patches for Security Gateways and Security Management as soon as possible.
• Verify which of your deployments fall into the affected product sets before assuming you are in the clear.
• Monitor vendor advisories for updated technical detail, since the triggering conditions remain undisclosed.

Source: The Hacker News

Are your Check Point gateways or management servers already patched against these two flaws?

This week's security roundup keeps circling back to one uncomfortable question: how did any of this get through? An extension asks for more access than it needs and gets it. A trusted service quietly becomes a link in a phishing chain. An old bug still delivers results. An exposed system stays exposed. A package looks legitimate right up until it isn't.

The common thread across these stories is that the way in rarely required anything sophisticated. 200 Android flaws headline the week's collection, alongside phishing campaigns assembled entirely inside the browser and a sprawling network of 119,000 scam shops. The roundup also bundles in 23 additional stories covering the broader threat landscape.

A few patterns worth noting from the collection:

• Extensions and packages that appear useful until they turn malicious
• Trusted services being abused as part of phishing infrastructure
• Long-patched vulnerabilities that remain effective because systems were never updated
• Exposed systems that simply stay exposed

The recurring lesson is that exposure often persists not because attackers are clever, but because basic hygiene gets skipped.

Source: The Hacker News

Which of these exposure patterns does your organization struggle with most, and how are you addressing it?

Identity was the primary target in roughly half of all confirmed malicious activity, according to Prophet Security, which investigated every alert across customer environments between May and July 2026. The analysis breaks down the four main attack patterns observed during that quarter and examines why some attacks succeeded while others were blocked.

The findings highlight that attackers continue to concentrate on identity-based techniques, making credential theft, session abuse, and authentication manipulation central to modern intrusion attempts. The report also contrasts successful compromises against blocked attempts, offering insight into which defensive controls proved effective and where gaps remain.

• Identity was the target in approximately 50% of confirmed malicious activity
• Data covers customer environments from May to July 2026
• Four main attack patterns were identified across the quarter
• The analysis explains why some attacks succeeded while others were blocked

The research focuses on the difference between detection and prevention, noting that investigating every alert provided a clearer picture of attacker behavior than isolated incident reviews. Organizations relying on identity-centric defenses should treat these patterns as a reminder that authentication boundaries remain a primary battleground.

The full breakdown of the four threat patterns and the reasons behind successful versus blocked attacks is available in the original report.

Source: BleepingComputer

How is your organization adapting its identity security controls in response to these findings?

IDScan has confirmed that attackers accessed customer data stored in its cloud platform, following reports that the company is linked to a leaked database containing more than 153 million driver's license scans.

The breach is tied to a massive trove of identity documents, and the confirmed access to customer data stored in the company's cloud environment indicates that the incident extends beyond a single exposed system.

Affected individuals and organizations should be aware of the following:

• Customer data stored in IDScan's cloud platform was accessed by attackers.
• The leaked database reportedly contains over 153 million driver's license scans.
• The company has confirmed the breach following earlier reports of the massive dataset.

Those potentially impacted should monitor for identity theft and phishing attempts that may leverage stolen license data.

Source: BleepingComputer

Has your organization reviewed its identity verification vendor relationships in light of this incident?

A new Android malware family called Mantax Otax is making the rounds, and it stands out because it blends ransomware and spyware functionality into a single payload. According to reporting, the malware is capable of encrypting files on the infected device, exfiltrating sensitive data, and then spamming and harassing victims, adding a psychological pressure layer on top of the usual data-loss impact.

This dual-purpose approach is notable for the Android threat landscape, where ransomware and spyware have typically been separate categories. Combining file encryption with data theft gives attackers two leverage points, and the added harassment component suggests the operators are prioritizing intimidation to push victims toward paying or complying.

Key facts to keep in mind:

• The malware is tracked as Mantax Otax
• Capabilities include file encryption, data theft, and victim harassment via spam
• It targets the Android platform

If you manage Android devices in your environment, it is worth reviewing your mobile threat defense posture, restricting sideloaded app installs where possible, and keeping an eye out for unusual file-access or encryption activity on endpoints.

Source: Unknown

Has anyone seen samples or indicators tied to Mantax Otax in the wild, and how is your team handling mobile ransomware exposure?

Surfshark has disclosed that hackers gained access to one of its internal test servers after a configuration error left the system exposed to the internet. The VPN provider says the incident was limited to internal testing and proxy servers, not its core VPN infrastructure or customer data.

According to the company, the misconfigured server was reachable from the public internet due to an error in its setup. Once inside, the attackers moved to a proxy server. Surfshark has not released specific details about how long the access lasted or exactly what data, if any, was viewed.

Key points from the disclosure:

• A configuration error exposed an internal test server to the internet.
• Attackers accessed that test server and a proxy server.
• The breach did not affect the core VPN service or customer accounts, per Surfshark.
• No specific CVE ID or advisory number has been provided in the report.

Users and administrators should watch for any follow-up guidance from Surfshark, especially if they operate their own proxy or test infrastructure that might be similarly exposed. As always, keep internal testing environments off the public internet unless absolutely necessary, and audit configurations regularly.

Source: Unknown

Has your organization reviewed whether any internal test or proxy servers are inadvertently exposed to the internet?

Cisco Talos has reported that two recently patched vulnerabilities in Secure Firewall Management Center (FMC) have been exploited in the wild by three distinct threat clusters. These clusters have been linked to both ransomware operations and state-sponsored cyber espionage activity, indicating the flaws are being actively weaponized by a range of adversaries.

The exploitation activity targets Cisco's Secure Firewall Management Center, a centralized management platform for Cisco firewall deployments. Organizations running affected FMC versions should prioritize patching, as the vulnerabilities have moved beyond theoretical risk into active exploitation by multiple groups with different motivations.

Details remain limited in the initial disclosure, but the involvement of three separate clusters suggests broad interest in the flaws across both criminal and nation-state actors. Administrators are advised to review Cisco's security advisories and apply the available fixes without delay.

• Apply the latest patches for Secure Firewall Management Center
• Review Cisco Talos guidance and advisories for affected versions
• Monitor for signs of exploitation in firewall management infrastructure

Source: BleepingComputer

Has your organization already patched its FMC deployments, or are you still assessing exposure?

A threat actor, described as likely Russian-speaking, leveraged hundreds of AI agents to build and run a global exploitation campaign against vulnerable PaperCut NG/MF servers, ultimately compromising 395 organizations.

The operation stands out less for any single novel flaw and more for how automation was used to scale reconnaissance, vulnerability discovery, and exploitation across a large number of targets. Rather than manually probing each victim, the actor reportedly relied on fleets of AI agents to handle repetitive tasks and accelerate the attack lifecycle.

Key points from the reporting:

• Target: internet-exposed PaperCut NG/MF print management servers
• Method: exploitation of known PaperCut vulnerabilities at scale
• Scale: 395 organizations impacted worldwide
• Attribution: likely Russian-speaking threat actor
• Notable element: use of hundreds of AI agents to develop and conduct the campaign

For defenders, the familiar fundamentals still apply: inventory any PaperCut NG/MF instances reachable from the internet, confirm they are patched against known vulnerabilities, and restrict exposure where printing infrastructure does not need to be publicly accessible. Given the automated nature of this campaign, delays in patching and asset tracking are exactly the gaps such tooling is designed to find.

Source: BleepingComputer

Does your organization run PaperCut NG/MF, and if so, is it exposed to the internet or kept internal-only?

Multiple cyber-espionage groups have been observed deploying an exploit kit tracked as BlueMoon, which chained zero-day vulnerabilities in Microsoft Windows and Google Chrome.

According to reporting, the kit was used in targeted campaigns, with the zero-days giving attackers a way into victim environments before patches were available. The activity has been attributed to more than one espionage-focused threat group, suggesting the tooling was shared or reused across operations.

Key points:

• The exploit kit is named BlueMoon.
• It leveraged zero-day flaws in Windows and Chrome.
• Multiple cyber-espionage groups were involved in deploying it.
• The campaign relied on previously unknown vulnerabilities, meaning no fix was available at the time of exploitation.

Details on specific affected versions, indicators of compromise, and mitigation guidance were not provided in the source material.

Source: BleepingComputer

Has your organization reviewed its Windows and Chrome patch cadence in light of shared exploit kits like this one?