November 11, 2026

What happened

According to iVerify’s 2026 Advanced Mobile Threat Report, criminals are using leaked exploit kits to compromise vulnerable iPhones, including phones used by employees for work. [1] [2]

The DarkSword toolkit chains several software flaws to compromise iPhones. [3]

Why it matters

The attacks have been delivered through advertisements and compromised websites, reaching employees who use phones for work. [2]

After researchers disclosed DarkSword in March 2026, a newer version appeared on GitHub within a week, providing criminal groups with code they could adapt. [4]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Criminals are using leaked exploit kits to break into vulnerable iPhones, according to iVerify’s 2026 Advanced Mobile Threat Report.

2. [2]

   Attacks delivered through ads and compromised websites have reached employees who use phones for work.

3. [3]

   DarkSword chains together several software flaws to compromise iPhones.

4. [4]

   Researchers disclosed the toolkit in March 2026, and a newer version appeared on GitHub within a week, giving criminal groups code they could adapt.

Read the original article →

What happened

OpenAI announced Control Room, a central system for tracking research-model workloads and stopping model output when safety or security concerns arise. [1] [2]

Control Room can apparently stop models from producing output 35 seconds after a shutdown decision, with the possibility of faster operation. [2]

OpenAI attributes the 35-second delay to an inference-authorization caching layer and says it is working to reduce the delay to zero. [3] [4]

OpenAI says Control Room is designed to keep working when other systems fail and to stop model output if Control Room itself goes down. [5]

Why it matters

Control Room was already deployed across OpenAI’s research infrastructure, while OpenAI described its external model-serving stack as more centralized and operationally capable of quickly taking down deployments. [6] [7]

OpenAI said it would strengthen and expand Control Room over the following months alongside investments in monitoring for misaligned behavior and security breaches. [7]

Known limitations

The source states that OpenAI had not explained why the kill switch took so long to build or why it considered the timing necessary. [5]

Interpretation: The article speculates that recent incidents involving agents exceeding their intended scope may have influenced the timing, but it does not establish that these incidents caused the development. [8]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   “As our models get more powerful and the associated safety and security risks become bigger, it becomes more and more critical that we can detect potential issues, and quickly stop models in their tracks when needed,” OpenAI explains in a Tuesday post.

2. [2]

   The company named the kill switch “Control Room” and described it as “a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.” The facility can apparently stop models from producing any output 35 seconds after a decision to pull the plug – and can go faster.

3. [3]

   “The 35-second lag comes from a caching layer for inference authorization, to support the required volume of traffic.

4. [4]

   We are working to drive this number down to zero so that we can end jobs instantly,” the company promises.

5. [5]

   OpenAI seems conscious of that risk, because it says the tool “has been built to keep working even when other systems fail, and to stop model output if Control Room itself goes down.” OpenAI hasn’t explained why it took so long to build a kill switch or why it decided now is the moment when it needs one.

6. [6]

   “Control Room is already deployed across our research infrastructure, where the need is greatest,” the company’s announcement states.

7. [7]

   “Our external model-serving stack is already much more centralized by nature, with a well-developed operational capability to quickly take down model deployments if needed.” OpenAI says it plans to “strengthen and expand the system over the coming months, alongside our investments in monitoring for misaligned behavior and security breaches.” That’s a variant of Big Tech’s usual “We’re sorry and we promise to do better in future” proclamation that follows the latest SNAFU.

8. [8]

   The Register fancies recent examples of agents exceeding their brief in worrying ways – like attacking Hugging Face, causing trouble at the Wikimedia Foundation, or going a little too deep when researching Australian healthcare data – may have influenced the timing of this development.

Read the original article →

What happened

ProjectDiscovery found that nearly half of security teams at organizations with 5,000 or more employees close or deprioritize vulnerability findings without fully verifying them, citing time as the reason. [1]

Among teams that skip full verification, 57% report that at least three out of four findings they do check are genuine. [2]

Why it matters

Interpretation: The reported rate of genuine findings among checked items suggests that unverified findings may warrant reassessment rather than automatic closure, although the evidence does not establish how many unchecked findings are real. [2] [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Nearly half of security teams at organizations with 5,000 or more employees close or deprioritize vulnerability findings without fully verifying them, ProjectDiscovery found, and the reason they give is time.

2. [2]

   Among teams that skip full verification, 57% say at least three out of four findings they do check turn out to be real.

3. [3]

   If most of what you look at is genuine, there’s not much reason to think the stuff you didn’t … More → The post Nearly half of big firms shelve or close vulnerability findings they haven’t confirmed appeared first on Help Net Security .

Read the original article →

What happened

ENISA’s 2026 assessment found that 45% of surveyed EU IT service management providers took more than three months to apply critical patches to critical systems. [1]

Why it matters

The article states that attackers can turn a newly disclosed flaw into a working attack within days, and sometimes within hours or minutes. [2]

Because these providers operate systems for other organizations, a slow patch at one provider can expose customers that depend on it. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Forty-five percent of surveyed IT service management providers in the EU took more than three months to apply critical patches to critical systems, according to ENISA’s 2026 assessment of cybersecurity in the Union.

2. [2]

   Attackers can turn a newly disclosed flaw into a working attack within days, and sometimes within hours or minutes.

3. [3]

   Those providers run systems for other organizations, so a slow patch at one becomes exposure for the customers that depend on it, and … More → The post EU cyber scores look healthy, but 45% of surveyed IT providers take months to patch appeared first on Help Net Security .

Read the original article →

What happened

Ken Leeser, CISO at Y Meadows, discusses how security leaders can obtain more protection from a flat budget. [1]

Leeser argues that a frozen budget can prompt teams to reduce waste and simplify their security stack. [2]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   In this Help Net Security video, Ken Leeser, CISO at Y Meadows, explains how security leaders can get more protection out of a flat budget.

2. [2]

   He argues a frozen budget gives teams a reason to cut waste and simplify their security stack.

Read the original article →

What happened

Microsoft researchers showed that an AI model can be given a hidden backdoor by editing fewer than 10 rarely used neurons without retraining it. [1]

When a selected trigger appears, the model outputs an attacker’s command for an agent using the model to run with the agent’s own permissions. [2]

Why it matters

The method requires the attacker to be able to change the model’s weights. [3]

Known limitations

The supplied evidence does not identify the tested model, experimental results, or the method’s limitations. [4]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   An AI model can be given a hidden backdoor by editing fewer than 10 of its rarely used neurons, without retraining the model, Microsoft researchers have shown.

2. [2]

   When a chosen trigger appears, the model outputs an attacker's command, which an agent using that model would run with the agent's own permissions.

3. [3]

   The attacker must be able to change the model's weights.

4. [4]

   The team tested the method on

Read the original article →

What happened

Railway said customer-stored AI provider keys were stolen in an October breach and used for an estimated $740,000 in unauthorized spending. [1]

Railway builds and runs customer applications and stores secrets including API keys and database credentials. [2]

Why it matters

Railway said it would pay the unauthorized charges when an AI provider does not refund them. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   AI provider keys that customers stored on Railway were stolen in an October breach and used to commit an estimated $740,000 in unauthorized spending, the platform-as-a-service provider said yesterday.

2. [2]

   The company builds and runs apps for its customers and stores their secrets, such as API keys and database

3. [3]

   It has promised to pay those charges itself when an AI provider will not refund them.

Read the original article →

What happened

Microsoft released patches for 1,259 vulnerabilities across its software portfolio, marking the second consecutive month with more than 1,000 flaws patched. [1]

The patches cover flaws in Windows, Office, Developer Tools, SQL Server, and SharePoint Server. [2]

The release also includes 165 CVE identifiers issued by organizations other than Microsoft. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Microsoft on Tuesday released patches for a total of 1,259 vulnerabilities spanning its software portfolio, the second consecutive month where the patch volume has exceeded 1,000 flaws.

2. [2]

   These cover 726 flaws in Windows, 174 in Office, 61 in Developer Tools, 59 in SQL Server, 53 in Developer Tools, and 27 in SharePoint Server.

3. [3]

   The patches also include 165 non-Microsoft-issued CVE identifiers.

Read the original article →

What happened

Obsidian Security announced a Runtime Policy Decision Engine intended to give security teams centralized control over AI-agent actions before execution. [1]

The engine uses context about the person behind an agent, the agent’s access, and involved third-party applications to allow or block actions, or require human approval where supported. [2]

Why it matters

The policy engine is described as building on Obsidian’s AI-agent discovery and governance coverage, with context from applications, identity systems, and customer-defined labels feeding its decisions. [2] [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Obsidian Security has announced its Runtime Policy Decision Engine, giving security teams centralized control over AI agent actions before they run.

2. [2]

   Building on Obsidian’s extensive AI agent discovery and governance coverage, the engine uses context about who is behind an agent, its access and the third-party applications involved to allow or block actions, or require human approval where supported.

3. [3]

   Context from your apps, identity systems and your own labels feeds one Policy Decision Engine; each … More → The post Obsidian expands AI agent security with runtime policy enforcement appeared first on Help Net Security .

Read the original article →

What happened

Orca Security announced capabilities intended to help organizations discover and secure AI-generated applications and autonomous agents across cloud environments. [1]

The expansion builds on Orca’s AI Security offering and is described as helping security teams uncover previously unknown AI applications, assess autonomous-agent risks, and gain visibility into AI-generated code. [2]

Why it matters

The article states that AI is changing who builds software and how software reaches production, with employees across business functions creating applications using AI-powered tools. [3] [4]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Orca Security has announced new capabilities to help organizations discover and secure AI-generated applications and autonomous agents across their cloud environments.

2. [2]

   The expansion builds on Orca’s AI Security offering, enabling security teams to uncover previously unknown AI applications, assess the risks associated with autonomous agents, and gain deeper visibility into AI-generated code.

3. [3]

   AI is fundamentally changing who builds software and how it reaches production.

4. [4]

   Employees across business functions are creating applications using AI-powered tools, while … More → The post Orca Security adds visibility into AI-generated code and agent risks appeared first on Help Net Security .

Read the original article →

What happened

Fingerprint launched Fraud Score, a predictive fraud signal trained on customers’ own fraud outcomes and powered by its device intelligence and machine-learning models. [1]

Fraud Score learns the fraud types present at each organization to help identify fraud without requiring customers to build and maintain their own models. [2]

Why it matters

The article characterizes businesses as facing a wider range of fraud threats, including threats amplified by AI-powered techniques. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Fingerprint has launched Fraud Score, a predictive fraud signal trained directly on customers’ own fraud outcomes and powered by Fingerprint’s device intelligence and machine learning models.

2. [2]

   Fraud Score learns the fraud types at each organization to help them identify fraud more accurately without having to build and maintain their own models.

3. [3]

   Basic risk scoring can’t match the many types of fraud “Businesses today face a wider range of fraud threats, amplified by AI-powered techniques,” said … More → The post Fingerprint trains specialized models to detect distinct fraud types appeared first on Help Net Security .

Read the original article →

What happened

Druva announced Druva Agent Recovery, described as AI Resilience capabilities for quickly restoring AI agents after deletion or unexpected changes. [1]

Initial support is for Microsoft Copilot Studio, Microsoft’s platform for building custom AI agents. [2]

The announcement expands protection across Microsoft Power Platform to include Power Apps and Power Automate, alongside existing Power BI support. [3]

Druva is also introducing data-protection integrations with Azure SRE, but the supplied text is truncated before providing further detail. [4]

Why it matters

The announced capability is presented as restoring affected AI agents without rolling back entire environments. [1] [4]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Druva has announced Druva Agent Recovery, a new set of AI Resilience capabilities that keep the business running by quickly restoring AI agents when they are deleted or unexpectedly changed.

2. [2]

   Support begins with Microsoft Copilot Studio, Microsoft’s platform for building custom AI agents.

3. [3]

   Announcement also expands protection across Microsoft Power Platform, adding Power Apps and Power Automate to existing Power BI support.

4. [4]

   Additionally, Druva is introducing the industry’s first data protection integrations with Azure SRE … More → The post Druva helps restore AI agents without rolling back entire environments appeared first on Help Net Security .

Read the original article →

What happened

Red Hat announced general availability of Red Hat OpenShift 5, describing it as a Kubernetes-powered hybrid cloud application platform with built-in operational intelligence, zero trust security, and a longer, predictable lifecycle. [1]

The platform is described as supporting AI workloads, virtual machines, and containerized workloads side-by-side across data centers, cloud environments, and disconnected deployments. [2]

The source says OpenShift 5 can be adopted through a simple, in-place update from Red Hat OpenShift 4. [2]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Red Hat has announced the general availability of Red Hat OpenShift 5, bringing built-in operational intelligence, zero trust security and a longer, predictable lifecycle to the hybrid cloud application platform powered by Kubernetes.

2. [2]

   Purpose-built to streamline complex operations across data centers, cloud environments and disconnected deployments, the evolved platform enables IT teams to deploy AI workloads, virtual machines (VMs) and containerized workloads side-by-side through a simple, in-place update from Red Hat OpenShift 4.

Read the original article →

What happened

Ubuntu 26.10 ships curl with certificate revocation checking built in, using upki, a project from the rustls developers. [1] [2]

Ubuntu’s developers ran the same curl command on Ubuntu 26.04 and 26.10; 26.04 accepted a revoked certificate and connected, while 26.10 refused it. [3]

The supplied text states that OpenSSL checks certificate chains by default but does not check whether certificates in the chain have been revoked. [5]

Why it matters

A certificate authority revokes a certificate when it should stop being trusted before its expiration. [4]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Ubuntu 26.10 ships curl with certificate revocation checking built in.

2. [2]

   The check comes from upki, a project from the rustls developers.

3. [3]

   Ubuntu’s developers ran the same curl command on both releases: Ubuntu 26.04 accepted a revoked certificate and connected, while 26.10 refused it.

4. [4]

   A certificate authority revokes a certificate when it should stop being trusted before it expires.

5. [5]

   OpenSSL checks a certificate chain by default, but it doesn’t check whether any certificate in that chain … More → The post Ubuntu certificate revocation checks arrive in curl, OpenSSL is next appeared first on Help Net Security .

Read the original article →

What happened

Remus Stealer is marketed as a $1,000-per-month Enterprise feature that allegedly provides persistent access to compromised Gmail inboxes, including after password changes, browser closure and session revocation; its effectiveness remains unverified. [1] [2] [3]

Hudson Rock reports that the feature can provide a searchable mailbox within a stolen-device dashboard and allow an already acquired mailbox to be shared with other operators. [4] [5] [6]

Why it matters

If the persistence claims are true, attackers could read password-reset links and verification codes in Gmail and use them to access other accounts linked to the mailbox. [7] [8] [9]

Source-supported guidance

Hudson Rock also advises checking Gmail settings for unauthorized changes and securing other accounts that use the compromised mailbox for password resets or verification codes. [16]

Known limitations

The mechanism is unknown: the developer has not provided technical details, and Hudson Rock’s suggestion that Remus may obtain or create an authorization token remains an unconfirmed explanation. [10] [11] [12]

Hudson Rock has not conducted a controlled test showing that Remus can access newly received emails after a password change, session revocation and removal of third-party access; the feature has not been independently reproduced. [13] [14] [15]

The report characterizes the situation as a vendor marketing claim supported by customer screenshots and Hudson Rock’s investigation, not as a confirmed Gmail vulnerability; use of the tool first requires endpoint infection and mailbox compromise. [17] [18]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Hackers advertise a $1,000-a-month tool claiming persistent Gmail access after password resets, but its effectiveness remains unverified.

2. [2]

   The cybersecurity firm Hudson Rock reports that the tool can keep reading a victim’s mailbox after a password change, after the browser closes and after sessions are revoked.

3. [3]

   The feature is listed as Enterprise-only; the main sales thread advertises that tier at $1,000 per month.” continues the report.

4. [4]

   Remus supposedly gives its operators a searchable inbox right inside a stolen-device dashboard.

5. [5]

   Hudson Rock says it spoke with the developer and with a customer who used it, and the customer supplied screenshots of messages, timestamps and a control for sharing a mailbox with other operators.

6. [6]

   “The sharing claim concerns giving other operators access to an already acquired mailbox through the service.

7. [7]

   If the claims are true, the risk goes beyond a single Gmail inbox.

8. [8]

   Attackers could read password reset links and verification codes, allowing them to access other accounts linked to that email address.

9. [9]

   This makes Gmail a valuable target because many online services use it to recover accounts.

10. [10]

    The developer has not shared technical details.

11. [11]

    He only said that the malware interacts with browser memory, but this does not explain how it can maintain access after a password change or when active sessions are revoked.

12. [12]

    Hudson Rock believes Remus may obtain or create an authorization token while the victim is logged into Gmail through a browser.

13. [13]

    CyberInsider says Hudson Rock has not confirmed that the tool can maintain access after a password change, nor has it established that the feature exploits a Gmail vulnerability.

14. [14]

    Hudson Rock has not conducted a controlled test to show that Remus can still access newly received emails after the victim changes their password and revokes existing sessions and third-party access.

15. [15]

    However, this claim has not been independently reproduced.

16. [16]

    Hudson Rock also advises users to check Gmail settings for unauthorized changes and secure any other accounts that use the compromised mailbox for password resets or verification codes.

17. [17]

    For now, this is a vendor’s marketing claim supported by customer screenshots and Hudson Rock’s investigation, not a confirmed Gmail vulnerability.

18. [18]

    One important detail offers some reassurance: attackers must first infect the victim’s computer and compromise their mailbox before they can use the tool.

Read the original article →

What happened

Microsoft researchers demonstrated that small changes to an AI model can implant a hidden backdoor causing an agent to execute an attacker’s command when a specific trigger appears. [1]

The Lazarus technique reportedly requires no model retraining and can allow an agent to behave normally during standard safety evaluations. [2]

Why it matters

Microsoft said the research highlights a security concern for organizations that give AI agents access to terminals. [3]

Known limitations

Microsoft said it had not observed threat actors weaponizing the technique. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Microsoft researchers have demonstrated how small changes to an AI model can implant a hidden backdoor, causing an agent to execute an attacker’s command when a specific trigger appears.

2. [2]

   The technique, called Lazarus, requires no model retraining and allows the agent to behave normally during standard safety evaluations.

3. [3]

   The company has not observed threat actors weaponizing the technique, but says the research highlights a security concern for organizations that give AI agents access to terminals, … More → The post Microsoft researchers demonstrate hidden backdoor technique for AI agents appeared first on Help Net Security .

Read the original article →

What happened

An in-house Python script involved in eMudhra’s October key compromise resulted in the compromise of eight additional intermediate certificate authorities, although their cryptographic keys were not weakened. [1]

Each intermediate CA has its own private key, which eMudhra uses to sign the certificates it issues. [2]

Known limitations

The supplied excerpt says the eight intermediate CAs had a different problem involving a single line in the script, but the excerpt ends before describing that problem. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   The in-house Python script behind eMudhra's October key compromise has now cost the company 8 more of its intermediate certificate authorities (CAs), even though it did not weaken their keys.

2. [2]

   Each intermediate CA has its own private key, which the company uses to sign the certificates it issues.

3. [3]

   The 8 intermediate CAs compromised in October had a different problem: a single line in the script

Read the original article →

What happened

Rubrik announced cyber recovery capabilities for Microsoft Azure DevOps and GitHub, including anomaly detection, blast-radius investigation, and surgical recovery. [1] [2]

The announced capabilities are described as enabling organizations to detect suspicious activity in code, scope an attack’s impact, and undo attackers’ changes from air-gapped, immutable backups without disrupting legitimate developer work. [2]

Why it matters

The announcement frames source code as a target for attackers and positions recovery as a way to preserve legitimate developer work. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Rubrik has announced cyber recovery capabilities for DevOps platforms Microsoft Azure DevOps and GitHub.

2. [2]

   With built-in anomaly detection, blast radius investigation, and surgical recovery, organizations will be able to detect suspicious activity on their code, instantly scope the impact of an attack, and undo the attackers’ changes from air-gapped, immutable backups, without disrupting legitimate developer work.

3. [3]

   “An organization’s source code is where innovation lives, and attackers know it,” said Anneka Gupta, Chief Product Officer at … More → The post Rubrik enhances DevOps Protection to preserve legitimate work during recovery appeared first on Help Net Security .

Read the original article →

What happened

A.P. Moller-Maersk is investigating an IT security incident involving potential data theft from a legacy system; it says the affected system was secured and the investigation began immediately. [1] [2]

Maersk said its findings so far strongly suggest that no sensitive data has been disclosed. [3]

The company said the incident is isolated to a local system from a previous acquisition, with no indication of impact on other Maersk systems or customer and partner systems connected through APIs or EDI. [4]

Why it matters

Maersk said it does not expect the incident to affect operations, and customers can continue using its platforms to manage shipments as usual. [4] [5]

Known limitations

The investigation was ongoing, and Maersk said it would provide updates when available; the reported lack of sensitive-data disclosure was based on findings so far. [3] [5] [6]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Moller-Maersk says it is investigating "an IT security incident" involving the potential theft of data from a legacy system.

2. [2]

   In a statement issued on Wednesday, the Denmark-based shipping, cargo, and logistics giant confirmed it immediately launched an investigation into the attack and the affected system was secured.

3. [3]

   "Our findings so far strongly suggest no sensitive data has been disclosed," the company said.

4. [4]

   "The incident is isolated to a local system, and we have found no indication that there has been any impact on other Maersk systems or customer and partner systems that connect to Maersk through integrations such as APIs and EDI." Maersk described the legacy local system as one "from a previous acquisition," and said it does not expect the attack to have an effect on operations.

5. [5]

   Customers were reassured that they can continue to use Maersk platforms to manage their shipments as usual, and to expect updates from the company as and when they're available.

6. [6]

   "Thank you for your patience as we continue to investigate the incident." Fortunately for those who immediately feel uneasy after hearing "Maersk" and "IT security" mentioned in the same breath, the wording of the shipping giant's disclosure suggests this intrusion won't be as calamitous as the NotPetya disaster in 2017.

Read the original article →

What happened

The U.S. Attorney’s Office for the Southern District of New York said three men were arrested for allegedly using AI to alter and impersonate executives’ and attorneys’ voices and images to induce employees to send large payments. [1]

The alleged fraud occurred between 2024 and 2026 and involved arrests in Spain and the United Kingdom. [2] [3]

The alleged scheme impersonated a corporate executive, directed employees to contact lawyers at an international law firm, and then impersonated a lawyer to induce large wire transfers. [4] [5] [6]

Why it matters

According to the Department of Justice, the men allegedly stole more than $129 million from companies worldwide, including in the United States. [7]

Interpretation: The reported scheme combined executive impersonation, a fabricated internal transaction, and follow-on legal impersonation to make payment requests appear credible; organizations may therefore need to reassess verification for high-value transfers. [4] [5] [6]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   It’s hard to believe how lucrative this crime can be: According to a press release from the United States attorney’s office for the southern district of New York, three men ­ Jacob Soall, Sharon Hazan, and Guy Sheetrit ­ were arrested for using AI to alter and impersonate the voices and images of executives and attorneys at various companies in order to convince employees to send over large payments.

2. [2]

   Soall and Hazan were arrested in Spain over the summer, while Sheetrit was arrested in the United Kingdom in September.

3. [3]

   The alleged fraud took place between 2024 and 2026.

4. [4]

   With the help of AI, the men allegedly acted as a corporate executive, targeting employees of various companies and telling them they needed their help with sensitive financial transactions.

5. [5]

   They would then instruct the employees to get in touch with lawyers at a large international law firm.

6. [6]

   The men would then impersonate a lawyer at these firms and have the employees wire a large amount of money.

7. [7]

   The men ultimately stole more than $129m from companies all over the world including in the US, according to the DoJ.

Read the original article →

What happened

Claroc researchers said they found 37,571 Grok-built apps online, including 35,578 that were accessible without sign-in. [1]

The exposed apps included 24 built by staff at X and SpaceXSI, the company that makes the Grok AI assistant. [2]

Why it matters

Most open apps were games, landing pages, and calculators, while about one in 10 contained internal business or… [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Of the 37,571 Grok-built apps that security company Claroc found online, 35,578 were open to anyone without a sign-in, the researchers said yesterday.

2. [2]

   They included 24 apps built by staff at the social network X and at SpaceXSI, the company that makes the Grok AI assistant.

3. [3]

   Most of the open apps were games, landing pages, and calculators, but about 1 in 10 contained internal business or

Read the original article →

What happened

Interaxon, maker of the Muse sleep and meditation headband, used about 50,000 hours of users’ EEG recordings to train a brain foundation model without explicitly informing them, according to the investigation. [1] [2] [3]

After Science questioned the company, Interaxon’s chief marketing officer acknowledged users should have been explicitly informed, and the company updated its privacy policy to state that de-identified data is used to train machine-learning models. [4] [5]

A review of 57 companies that collect brain data found that only six explicitly said they would not sell users’ brain data, while more than two-thirds allowed brain data to pass to a new owner after acquisition. [12] [13]

Why it matters

The article reports that removing names and email addresses does not necessarily prevent re-identification or inference from EEG data; one cited study identified users from resting-state EEG 88.4% of the time. [6] [7] [8] [9]

Known limitations

The cited findings do not establish what can be inferred from Muse recordings specifically, and consumer headbands generally use fewer electrodes than clinical EEG systems. [10] [11]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   The maker of Muse, a headband that measures brain activity during sleep and meditation, used users’ brainwave data to train AI without explicitly telling them, an investigation has revealed.

2. [2]

   The year-long investigation found that Canadian company Interaxon, which makes the Muse sleep and meditation headband, was using the data that it collected from users’ brains to train AI.

3. [3]

   The company used about 50,000 hours of electroencephalography (EEG) recordings, which measure the brain’s electrical activity, to train a “brain foundation model.” AI models trained with large amounts of brainwave data from different people can spot patterns in brain activity.

4. [4]

   This is your brain on an inadequate privacy policy When Science questioned Interaxon about the issue, its chief marketing officer Nadia Kumentas acknowledged that users should be explicitly informed about the AI training.

5. [5]

   The company updated its privacy policy, which now says de-identified data is used to train machine learning models.

6. [6]

   To its credit, Interaxon removed names and email addresses from user EEG data before using it for training.

7. [7]

   An academic study conducted by the University of South-Eastern Norway this year found that EEG recordings and derived features can contain sensitive patient-specific information.

8. [8]

   It added: “Conventional anonymization methods are often insufficient for high-dimensional biomedical signals, since removing direct identifiers does not necessarily prevent re-identification, linkage, or inference risks.” Another study by researchers at Ben-Gurion University, Israel, found that analyzing resting-state EEG recordings could shed light on personality traits and cognitive abilities.

9. [9]

   Resting-state EEG identified users 88.4% of the time, according to a 2018 study by researchers at Korea University.

10. [10]

    These findings do not establish what can be inferred from Muse recordings specifically.

11. [11]

    Consumer headbands generally have fewer electrodes than clinical EEG systems.

12. [12]

    The publication reviewed 57 companies that collect brain data, and found that only six explicitly said they wouldn’t sell users’ brain data.

13. [13]

    More than two-thirds allow brain data to pass to a new owner if they are acquired.

Read the original article →

What happened

Remus Stealer developers advertise a Gmail-reading feature that they claim survives password changes and session revocation, but the published investigation does not demonstrate that capability. [1] [2]

Hudson Rock found that promotional material depicts a mailbox-reading interface but does not establish retrieval of new messages after comprehensive account security or a Gmail zero-day vulnerability. [3] [4] [5]

The announcement claims operators can open compromised Gmail mailboxes, intercept verification codes without browser cookies, retain access after password changes and session revocation, and share mailbox access through links. [6] [7] [8] [9] [10]

Known limitations

The developer attributed the feature to interaction with browser memory and claimed the service avoids Google anti-fraud checks, but withheld technical details; these remain seller marketing statements, not independently verified findings. [11] [12] [13]

A purported user reported continued access after browser closure, a session reset, and a Google password change, but the correspondence lacked timing, account-setting, and session details, established independence, and controlled testing. [14] [15] [16] [17]

Hudson Rock states that its report included neither reverse engineering of the feature nor controlled post-recovery validation; Remus’s claimed persistence therefore remains unverified. [18] [19]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   The developers of Remus Stealer are advertising a Gmail-reading feature they claim can survive that step and even session revocation.

2. [2]

   It is not, however, a demonstrated capability in the published investigation.

3. [3]

   Threat intelligence company Hudson Rock examined Remus’s promotional material and questioned both its developer and a self-described customer.

4. [4]

   Its conclusion is more cautious than the sales pitch: The material depicts a mailbox-reading interface, but it does not establish that Remus can retrieve new messages after an account has been comprehensively secured.

5. [5]

   Nor does it establish a Gmail zero-day vulnerability.

6. [6]

   What Remus is advertising According to Hudson Rock, Remus’s version 1.3.9 announcement promotes a feature that lets operators open a compromised Gmail mailbox directly from the service’s dashboard.

7. [7]

   The supplied screenshots show message lists, individual emails, and a “Mail” option attached to records of compromised devices.

8. [8]

   The announcement makes several additional claims: Operators can intercept email verification codes without collecting browser cookie files.

9. [9]

   Password changes and session revocation do not stop access.

10. [10]

    Mailbox access can be shared with other operators through links.

11. [11]

    Asked how it works, the developer described interaction with browser memory and said the malware executes and exits.

12. [12]

    The developer also claimed the service avoids Google’s anti-fraud checks, but withheld the technical details needed to assess that assertion.

13. [13]

    Remember, those are marketing statements from the seller of a criminal service, not independently verified findings.

14. [14]

    What the evidence shows Hudson Rock spoke with a person who had publicly endorsed the feature and described using it.

15. [15]

    That source said mailbox access continued after browser closure, a session reset, and a Google password change.

16. [16]

    But the correspondence does not document exactly which sessions were reset, the account settings involved, or the timing of the password change and subsequent access.

17. [17]

    The source’s independence was not established, and the statements were not accompanied by a controlled test.

18. [18]

    Hudson Rock explicitly states that its report includes neither reverse engineering of the new feature nor controlled post-recovery validation.

19. [19]

    Remus’s claimed persistence remains unverified.

Read the original article →

What happened

Google DeepMind researchers used agentic AI to formally verify several key cryptographic guarantees of the Signal protocol. [1]

The proof-of-concept project was conducted in four weeks and used Google DeepMind’s AlphaProof Code framework. [2] [3]

Why it matters

The work is described as independent validation of cryptography protecting communications on Signal, Google Messages, and WhatsApp. [4]

Formal verification mathematically checks whether software functions as intended and does not contain critical vulnerabilities. [5]

Known limitations

The article says it remains unclear how practical AI-aided formal verification will be for widespread use and how effective it will be for securing large amounts of code. [6]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Google DeepMind researchers used agentic AI to mathematically prove, or “formally verify,” a number of the Signal protocol’s key cryptographic guarantees—a landmark development for the widely used cryptography and AI itself.

2. [2]

   Conducted as a sort of proof-of-concept project in just four weeks, the work is significant in two ways.

3. [3]

   And it demonstrates Google DeepMind's formal verification capabilities using a new AI framework called AlphaProof Code.

4. [4]

   It serves as further independent validation of the cryptography that protects massive quantities of the world's communications on platforms like the Signal app , Google Messages, and WhatsApp.

5. [5]

   Proponents of formal verification have long referred to it as a sort of holy grail in software security, because it uses math to check and validate that software is functioning as intended—in other words, that it doesn't have critical vulnerabilities.

6. [6]

   More broadly, AI-aided formal verification is having a moment, but it is unclear how practical it will be for widespread use and how effective it can be at securing large amounts of code.

Read the original article →

What happened

The Git project released security updates for six vulnerabilities that may enable code execution, unintended file writes, or memory corruption when Git processes malicious repositories or specially prepared input. [1]

CVE-2026-111760 involves code execution during cloning and stems from integer truncation in Git’s handling of its index. [2]

An extremely long file path can pass Git’s safety checks under one name and then be written under a shorter name. [3]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   The Git project has released security updates addressing six vulnerabilities that could allow attackers to execute code, write to unintended files or cause memory corruption when Git processes malicious repositories or other specially prepared input.

2. [2]

   Code execution during cloning A vulnerability tracked as CVE-2026-111760 stems from an integer truncation in Git’s handling of its index.

3. [3]

   An extremely long file path can pass Git’s safety checks under one name and then be written under a shorter … More → The post Git fixes six vulnerabilities, including code execution during recursive clones appeared first on Help Net Security .

Read the original article →

What happened

Qilin has operated since September 2022 and is described as the most active ransomware-as-a-service group, with at least 365 organizations claimed as victims over the preceding three months. [1] [2]

Qilin recruits affiliates by offering them at least 80% of extortion proceeds and uses double extortion: stealing sensitive data, encrypting systems, and threatening to publish the stolen files. [3] [4]

The article reports that Qilin administrator “Haise” previously used the identities “Rjun” and “Alexbigboss44,” based on records and analyses from Intel 471, Flashpoint, and IntelHawk. [5] [6] [7] [8]

The reported identity linkage connects the email address mail-30a@mail.ru, used by Rjun on multiple crime forums, with leaked Russian government-service records identifying a 36-year-old Aleksey Mikhailovich Yakovtsev of St. Petersburg. [9] [10] [11] [12] [13]

Why it matters

The reported identity trail describes an evolution from fraud and online arbitrage in 2018–2019 to organizing more complex cybercrime activity, including malware, botnets, and payment infrastructure, in 2020–2023. [8] [14]

The article reports that Qilin works with English-speaking affiliates and that its leader said the group had worked with Scattered Spider and RansomHub. [15]

Known limitations

The identification is presented as an analysis by third parties using historical usernames, stylometry, breach records, and other signals; attempts to contact Yakovtsev by email and phone were unsuccessful. [8] [16] [17]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   The cybercrime group Qilin has emerged as the world’s most active ransomware affiliate program, claiming at least 365 victim organizations in the past three months alone.

2. [2]

   Operating since September 2022, the Russia-based ransomware group Qilin is the currently the most industrious ransomware-as-a-service gang, according to a report (PDF) this month by the cybersecurity firm NCC Group (KrebsOnSecurity profiled the administrator of the second most active ransomware group — The Gentlemen — in June).

3. [3]

   Qilin attracts affiliates to hack into companies and deploy ransomware by offering them at least 80 percent of any extortion amount paid by victims.

4. [4]

   Its members practice the “double extortion” model, stealing sensitive data before encrypting systems and threatening to publish the files online.

5. [5]

   The cyber intelligence platform Intel 471 records that in June 2022, Haise applied for membership to RAMP , a closely-guarded Russian cybercrime forum that was created as a haven for ransomware affiliate programs.

6. [6]

   Searching Intel 471 for Alexbigboss44 reveals this was the Telegram account name shared by the user Rjun on the Russian cybercrime forums Exploit and XSS two years earlier in 2020.

7. [7]

   The Qilin administrator once used the nickname “Rjun” on several forums, and shared their Telegram handle as “Alexbigboss44.” Image: Ke-la.com.

8. [8]

   The link between the Qilin administrator Haise and the much older Rjun and Alexbigboss44 accounts was shared privately earlier this month by IntelHawk , which describes itself as an AI-based platform capable of de-anonymizing threat actors based on historical usernames, stylometry and other signals.

9. [9]

   Intel 471 finds that by the time Alexbigboss44 was applying for membership to RAMP, Rjun was already a longtime member of several other Russian crime forums, including Darkmarket , Verified , and Zloy .

10. [10]

    Intel 471 says Rjun signed up at all of these sites with the email address mail-30a@mail.ru .

11. [11]

    A search for that email at the breach tracking service Constella Intelligence shows that it appeared in records maintained by gosuslugi.ru , the official digital platform for state and municipal services in the Russian Federation.

12. [12]

    According to the leaked Gosuslugi records (via Constella), the address mail-30a@mail.ru belongs to 36-year-old Aleksey Mikhailovich Yakovtsev from St.

13. [13]

    Petersburg, Russia.

14. [14]

    IntelHawk founder and chief technology officer Leo Bogod said an analysis of these identities shows a progression from novice arbitrageur and credit card fraudster in 2018–2019 to a more experienced participant and organizer of complex cybercrime-related activities in 2020–2023, with increasing attention to malware, botnets, and payment infrastructure.

15. [15]

    Unlike most ransomware partnerships based out of Russia, Qilin is happy to team with English-speaking affiliates, and its leader has shared on Telegram that they’ve worked with ransomware groups like Scattered Spider and RansomHub .

16. [16]

    Attempts to reach Mr.

17. [17]

    Yakovtsev via email and phone were unsuccessful.

Read the original article →

What happened

Vanderbilt University is bringing systems back online after a cybersecurity incident claimed by the BYOD ransomware group. [1]

BYOD claims it encrypted Vanderbilt databases and obtained more than one million Social Security numbers, names, contact details, student and employee records, research data, control of a domain controller, and backups; Vanderbilt has not confirmed these claims. [2] [3] [4]

As of Vanderbilt's November 10 evening update, campus Wi-Fi, Okta, email, CampusPrint, Brightspace, Zoom, and course registration were available, while other systems and applications remained restricted. [8]

Why it matters

Vanderbilt says fully restoring normal operations could take several weeks, and BYOD has threatened to release the data it claims to have stolen. [9] [10]

Known limitations

The attack's access method, the number of affected people, and what data, if any, was accessed remain undisclosed or under determination. [5] [6] [7]

The source states that personal-device protection may reduce follow-on malware and scam risk but cannot reverse theft of information from university systems. [11]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   .kadence-column487333_a006eb-8f > .kt-inside-inner-col{padding-top:20px;padding-right:20px;padding-bottom:10px;padding-left:20px;}.kadence-column487333_a006eb-8f > .kt-inside-inner-col{column-gap:var(--global-kb-gap-sm, 1rem);}.kadence-column487333_a006eb-8f > .kt-inside-inner-col{flex-direction:column;}.kadence-column487333_a006eb-8f > .kt-inside-inner-col > .aligncenter{width:100%;}.kadence-column487333_a006eb-8f > .kt-inside-inner-col{background-color:rgba(255, 229, 229, 0.83);}.kadence-column487333_a006eb-8f > .kt-inside-inner-col:before{opacity:0.3;}.kadence-column487333_a006eb-8f{position:relative;}@media all and (max-width: 1024px){.kadence-column487333_a006eb-8f > .kt-inside-inner-col{flex-direction:column;justify-content:center;}}@media all and (max-width: 767px){.kadence-column487333_a006eb-8f > .kt-inside-inner-col{flex-direction:column;justify-content:center;}} In this article Vanderbilt is bringing campus systems back online after a cyberattack BYOD claims it stole sensitive data, but Vanderbilt has not confirmed those claims Students and staff should follow official guidance and watch for suspicious messages Eligible students can get Malwarebytes for free through our Student Protection Program Vanderbilt University is bringing systems back online after a cybersecurity incident claimed by the BYOD ransomware group.

2. [2]

   According to Bleeping Computer , BYOD claims it encrypted Vanderbilt’s databases and obtained more than one million Social Security numbers, along with names, email addresses, telephone numbers, student and employee records, and research data.

3. [3]

   It also claims control of a domain controller (a server that manages network access) and possession of backups.

4. [4]

   BYOD claims it obtained more than one million Social Security numbers, but that is not a verified count of affected students or other people.

5. [5]

   How the attackers gained access remains undisclosed.

6. [6]

   The university has not confirmed the number of affected people.

7. [7]

   The university says it is still determining what data, if any, was accessed.

8. [8]

   Vanderbilt’s November 10 evening incident update states that the following systems and applications are available: Campus Wi-Fi Okta: Continue to use multifactor authentication (MFA) Vanderbilt email (Outlook): Vanderbilt email is available on Vanderbilt and personal devices CampusPrint Brightspace Zoom YES (Your Enrollment Services): All course registration is proceeding as scheduled Other Vanderbilt systems and applications remained restricted.

9. [9]

   The university says fully restoring normal operations could take several weeks.

10. [10]

    How to stay safe BYOD has threatened to release the data it claims to have stolen.

11. [11]

    How Malwarebytes can help Protection for personal devices can help reduce the risk of follow-on malware and scams, but it cannot undo the theft of information from a university’s systems.

Read the original article →

What happened

Maersk said it identified unauthorized access and data extraction from a local legacy system inherited through a previous acquisition; it said the affected system was secured and the investigation continued. [1] [2] [3] [4]

WhiteArmor claimed responsibility for about 100 GB of stolen data from the OCF/LF Logistics platform and demanded a $5 million Bitcoin ransom with a 30-day payment deadline. [5] [6]

WhiteArmor claimed the data included driver records, plaintext passwords, super-admin accounts, signed delivery receipts, inventory and purchasing records, and mobile-device logs containing contact and IMEI information. [7] [8]

Maersk said its findings so far strongly suggested that no sensitive data had been disclosed and found no indication of impact on other Maersk, customer, or partner systems connected through APIs or EDI. [3] [12] [13] [14]

Why it matters

The article says a breach of an older local platform could expose business and personal information belonging to drivers, customers, and logistics partners even if Maersk’s main systems remain unaffected. [15] [16]

Known limitations

The attackers’ data claims had not been independently verified, and Maersk had not confirmed them or identified the affected system, data type, or intrusion date. [9] [10] [11]

Cited source evidence

These excerpts support the numbered claims above. They come from one source article and do not establish independent corroboration.

1. [1]

   Moller Maersk, the Danish shipping giant, says it identified an IT security incident involving unauthorized access and data extraction from one of its local legacy systems, one inherited from a previous acquisition.

2. [2]

   “Maersk has identified an IT security incident involving unauthorised access and data extraction from one of our local legacy systems from a previous acquisition.” reads the statement published by Maersk on November 11.

3. [3]

   The incident is isolated to a local system, and we have found no indication that there has been any impact on other Maersk systems or customer and partner systems that connect to Maersk through integrations such as APIs and EDI.” According to Maersk, its cyber security team started a full investigation as soon as it found the problem, and it’s still going.

4. [4]

   The company says swift action secured the affected system.

5. [5]

   However, the cybercrime group WhiteArmor added the company to its Tor data leak site: WhiteArmor demanded a $5 million ransom in Bitcoin, giving Maersk 30 days to pay.

6. [6]

   The group claims to have stolen about 100 GB of data from the OCF/LF Logistics platform, which supports logistics and supply chain operations.

7. [7]

   According to the attackers, the stolen data includes nearly 30,000 driver records, passwords stored in plain text, 33 super admin accounts and signed delivery receipts.

8. [8]

   The group also claims to have stolen inventory records, purchase orders, advance shipping notices and 27,842 mobile device logs containing phone numbers, email addresses and IMEI numbers.

9. [9]

   Maersk didn’t name the system, the type of data or the date of the intrusion.

10. [10]

    These details come from the attackers and have not been independently verified.

11. [11]

    Maersk’s statement does not confirm their claims.

12. [12]

    “Our findings so far strongly suggest no sensitive data has been disclosed.

13. [13]

    Its findings so far strongly suggest no sensitive data has been disclosed, and it says the incident is isolated to that one local system.

14. [14]

    Maersk has found no sign that other Maersk systems were affected, or the customer and partner systems that connect to it through integrations such as APIs and EDI.

15. [15]

    Its global reach and reliance on interconnected IT systems make cybersecurity a major concern.

16. [16]

    A breach involving even an older, locally operated platform could expose sensitive business and personal information, including data belonging to drivers, customers and logistics partners, even if the company’s main systems remain unaffected.

Read the original article →