Google announced it will equip Chrome with an AI driven method to detect and block Tech Support Scam websites

Google has expressed plans to use Artificial Intelligence (AI) to stop tech support scams in Chrome.

With the launch of Chrome version 137, Google plans to use the on-device Gemini Nano large language model (LLM) to recognize and block tech support scams.

Users already have the ability to chose Enhanced Protection under **Settings > Privacy and security > Security > Safe Browsing.**

Safe Browsing settings

Google’s reasoning, and we agree, is that LLMs are fairly good at understanding and classifying the varied, complex nature of websites. Meaning that, since many malicious sites have a very short lifespan, it is more effective to learn and recognize their behavior rather than keep adding a host of domain names to a block-list (something which Google has frustrated with the introduction of Manifest V3, by the way).

Tech support scams typically follow a certain pattern that should be simple to learn:

*   They make your browser tab full screen
*   Display the number they want you to call all over the place
*   Show the visitor fake ongoing scans and alerts

These are just a few of the characteristics I’d teach the LLM. I’m not speaking for Google here. They just mention they’ll be looking at usage of the Keyboard Lock API.

On that, the Keyboard Lock API is a web technology that allows websites to “capture” keyboard input, meaning they can prevent certain key combinations (or all keys) from working as they normally do in your browser or operating system. Originally, this tool was designed for legitimate purposes, like making web games or remote desktop apps more immersive by stopping accidental key presses from interrupting the experience. But tech support scammers exploit the Keyboard Lock API to make it harder for victims to escape their scam pages. This means that when a visitor tries to close the scam page or switch to another program, nothing happens, making them feel trapped on the site. Which also makes them think their system is actually infected.

Google explains why it went for the on-device method, saying it allows them to see the threats at the same moment the users see them and in the same way the users see them.

> “We’ve found that the average malicious site exists for less than 10 minutes, so on-device protection allows us to detect and block attacks that haven’t been crawled before.”

**How it works**

When the user lands on a suspicious page, which is decided by the on-device LLM, based on specific triggers like the Keyboard Lock API, Chrome provides the LLM with the contents of the page that the user is on and queries it to extract security signals, such as the intent of the page. This information is then sent to a Safe Browsing server for a final verdict.

If Safe Browsing decides the website is malicious, Chrome will block the content and show the user a big warning screen, called an “interstitial.”

Image courtesy of Google

By making the target think their system is infected, tech support scammers try to gain remote access or obtain payment information. Google says:

> “Tech Support scams are an increasingly prevalent form of cybercrime, characterized by deceptive tactics aimed at extorting money or gaining unauthorized access to sensitive data.”

Malwarebytes’ Browser Guard data over the last month shows that 30% of the fraudulent websites we blocked through the browser extensions are tech support scams.

30% of the three fraud categories are TSS

So, it’s nice of Google to let Chrome help us take care of some of those, but Chrome is not the only browser. We’re even hearing stories from users that ran into a website telling them their Windows machine was infected while they were using the Safari browser on their iPad.

**We don’t just report on phone security—we provide it**

Cybersecurity risks should never spread beyond a headline. Keep threats off your mobile devices by downloading Malwarebytes for iOS, and Malwarebytes for Android today.

 Google Chrome will use AI to block tech support scam websites 

A flaw in Microsoft Entra ID’s legacy login allowed attackers to bypass MFA, targeting admin accounts across finance,…

A flaw in Microsoft Entra ID’s legacy login allowed attackers to bypass MFA, targeting admin accounts across finance, healthcare, and tech sectors.

Cybersecurity firm Guardz has discovered a targeted campaign exploiting a weakness in Microsoft Entra ID’s legacy authentication protocols, allowing attackers to bypass modern security measures like Multi-Factor Authentication (MFA).

The attacks, which occurred between March 18 and April 7, 2025, utilized Basic Authentication Version 2 – Resource Owner Password Credential (BAV2ROPC), a legacy login method, to gain unauthorized access, highlighting the dangers of outdated authentication in cloud environments.

This campaign, according to Guardz’s report, shared with Hackread.com, targeted various sectors including financial services, healthcare, manufacturing, and technology services.

This incident follows the widespread Microsoft Entra ID account lockouts reported by Hackread.com in April 2025, caused by an internal Microsoft error with refresh tokens and the MACE Credential Revocation app. While Hackread’s report detailed unintentional lockouts due to an internal issue, the Guardz discovery highlights a deliberate exploitation of Entra ID vulnerabilities by malicious actors.

****Campaign Analysis****

Guardz Research Unit (GRU) discovered that threat actors were actively exploiting BAV2ROPC – a compatibility feature within Entra ID that allows older applications to authenticate using simple usernames and passwords.

Unlike contemporary interactive login processes that mandate MFA and other security checks, BAV2ROPC operates non-interactively. This critical difference allows attackers to completely circumvent MFA, Conditional Access Policies, and even login alerts and user presence verification, effectively rendering these modern protections useless.

****Attack Timeline****

The attack occured in two distinct phases starting with an “Initialization” phase between March 18th and 20th, characterized by a lower intensity of probing, averaging around 2,709 suspicious login attempts daily.

This was followed by a “Sustained Attack” phase, from March 21st to April 3rd, which featured a dramatic surge in activity, spiking to over 6,444 attempts per day (a whopping 138% increase). This escalation indicated a clear shift towards aggressive exploitation of the identified vulnerabilities.

Guardz Research tracked over 9,000 suspicious Exchange login attempts, primarily from Eastern Europe and the Asia-Pacific region. The campaign involved automated credential spraying and brute-force tactics, focusing on exposed legacy endpoints.

The attacks targeted various legacy authentication vectors, with over 90% aimed at Exchange Online and the Microsoft Authentication Library, including a significant focus on administrator accounts.

“Admin accounts were a specific focus. One subset received nearly 10,000 attempts from 432 IPs within 8 hours,_“_ wrote Guardz’s Elli Shlomo in their blog post.

While the campaign has subsided, Guardz warns that the vulnerability persists in many organizations still relying on protocols like BAV2ROPC, SMTP AUTH, POP3, and IMAP4 for compatibility. These methods bypass MFA, ignore Conditional Access, and enable silent, non-interactive logins, thus, creating a “hidden backdoor,” researchers noted.

Dor Eisner, CEO and Co-Founder of Guardz emphasized the critical nature of this issue, stating, “This campaign is a wake-up call, not just about one vulnerability, but about the broader need to retire outdated technologies that no longer serve today’s threat landscape.”

To mitigate the risks, Guardz urges organizations to immediately audit and disable legacy authentication, enforce modern authentication with MFA, implement conditional access policies to block unsupported flows, and closely monitor for unusual login activity.

Legacy Login in Microsoft Entra ID Exploited to Breach Cloud Accounts

Cybersecurity researchers have flagged three malicious npm packages that are designed to target the Apple macOS version of Cursor, a popular artificial intelligence (AI)-powered source code editor.
"Disguised as developer tools offering 'the cheapest Cursor API,' these packages steal user credentials, fetch an encrypted payload from threat actor-controlled infrastructure, overwrite Cursor's

Malicious npm Packages Infect 3,200+ Cursor Users With Backdoor, Steal Credentials

Cloud-native applications offer scalable, automated workflows, intelligent data processing, and seamless deployments. However, many organizations still struggle to…

Cloud-native applications offer scalable, automated workflows, intelligent data processing, and seamless deployments. However, many organizations still struggle to manage their workflows effectively. Beneath polished interfaces and advanced features, many systems rely on scattered scripts, manual processes, and fragile pipelines that fail under pressure.

When I first encountered the scale challenges in cloud-native applications over 15 years ago, I was struck by the paradox: cloud systems promise efficiency and scalability, but often, organizations struggle under the weight of fragmented, inefficient workflows. That moment pushed me to find better solutions, and today, I’m excited to share some insights I’ve gathered along the way. 

I’m Aditya Bhatia, and in my experience leading cloud-native architectures, I’ve faced firsthand the hurdles organizations encounter when orchestrating workflows at scale. From building distributed orchestration systems to automating complex workflows with Kubernetes, I’ve learned how inefficient workflows don’t just harm operations, they inflate costs and put teams in a constant firefighting mode.

These problems are not merely technical hiccups; they stem from deeper architectural flaws where complexity overwhelms control. Many cloud workflows fail to scale under increased load, become cost-inefficient, or lack the resilience required for mission-critical operations. In this article, I’ll explore how mastering workflow orchestration, particularly through Kubernetes, can address these challenges and deliver a sustainable solution.

I’ll share insights from my experience with Kubernetes-based workflow orchestration, detailing key architectural patterns, best practices, and real-world examples. Whether managing complex data pipelines, building machine learning workflows, or maintaining mission-critical systems, you’ll learn how to design scalable, resilient workflows that drive cloud-native success.

****Understanding Workflow Orchestration Systems****

Workflow orchestration is more than automating processes, it’s about creating intelligent, scalable systems that streamline execution across distributed infrastructures. It ensures consistency, scalability, and efficiency, making it essential for cloud-native environments.

****Types of Workflows****

*   **Stateless Workflows:** These tasks do not maintain data between executions, making them ideal for scalable microservices and API-driven processes. For example, an API gateway that forwards user requests to different services without maintaining session data is stateless.

*   **Stateful Workflows:** These maintain data between executions and are critical for long-running tasks like machine learning pipelines, complex data processing, or multi-step transaction systems.

In my experience leading large-scale workflow automation, well-architected orchestration systems play a crucial role. Whether automating AI model training pipelines or enhancing the resilience of distributed services, orchestration forms the backbone of cloud-native infrastructures. 

Think of Kubernetes as the brain of your workflow system, it makes decisions about where and how things run, ensuring everything stays smooth even as demand fluctuates. Kubernetes simplifies these complexities by automatically adjusting to the workload, scaling seamlessly, and ensuring that resources are allocated exactly where needed, keeping your system reliable and efficient.

Research shows that Kubernetes is now a leading platform for managing cloud-native scientific workflows due to its scalability and flexibility. Similarly, industry reports highlight how Kubernetes simplifies CI/CD pipelines, solidifying it as an essential tool for workflow automation.

****The Architecture of Kubernetes-Based Workflow Orchestration****

Kubernetes is ideally suited for workflow orchestration due to its distributed, resilient, and scalable architecture. At its core, Kubernetes leverages the following components to manage workflows:

*   **Control Plane:** Manages the orchestration process, including the API Server, Scheduler, and Controller Manager, ensuring smooth coordination across the cluster.

*   **Worker Nodes:** These nodes execute workloads in containers, enabling seamless scaling as demand fluctuates.

*   **Operators and Custom Resource Definitions (CRDs):** Extend Kubernetes’ capabilities, automating complex, multi-step processes without manual intervention, thereby reducing overhead and error-prone tasks.

In my projects, I’ve designed orchestration systems that harness Kubernetes’ flexibility to manage and scale workflows. For example, KubeAdaptor integrates containerized workflows into Kubernetes, offering scalability, resource optimization, and simplifying orchestration management while ensuring high availability and performance.

To better understand Kubernetes-based orchestration. The diagram shows that KubeAdaptor integrates containerized workflows within the Kubernetes environment, streamlining resource management and ensuring scalability across the infrastructure.

****Adaptive Resource Management in Workflow Orchestration****

Scaling workflows presents significant challenges in resource management. Without effective allocation, workflows become unreliable and cost-prohibitive. Kubernetes’ dynamic resource management capabilities, particularly the **MAPE-K model** (Monitor, Analyze, Plan, Execute, Knowledge), address these challenges by optimally allocating resources to maintain performance and reduce infrastructure costs.

The **MAPE-K Model** enables Kubernetes to monitor workloads in real-time, adjust resources as necessary, and execute changes dynamically, ensuring that cloud infrastructure is used efficiently. By automatically aligning resources with workflow demands, Kubernetes saves time and money while maintaining system performance.

I remember one case where Flyte, a Kubernetes-native workflow engine, played a pivotal role in Freenome’s cancer detection research. The challenge was clear: they needed scalable workflow management that could handle the complexity of scientific research without being bogged down by resource limitations.

Using Kubernetes, we saw the system dynamically allocate resources based on real-time demand, giving them the needed performance boost, especially in a cloud environment where multiple teams share resources. It was a game-changer, turning what would have been a costly and inefficient process into a streamlined, high-performing solution.

****Scalable Workflow Management: The Worker Pool Model****

Scalability is a non-negotiable requirement in cloud-based workflow management. Kubernetes excels with the **Worker Pool Model**, which dynamically adjusts the number of workers based on demand, ensuring optimal resource allocation.

This model is especially valuable for cloud-native applications that require seamless scaling without manual intervention. Leveraging the Worker Pool Model, I’ve optimized resource utilization, scaling workers dynamically based on the complexity of incoming tasks. This ensures that workflows always run at peak efficiency, regardless of the workload’s size or unpredictability.

This approach is efficient in scientific workflows, where large datasets are processed and the demand for compute resources can fluctuate rapidly.

****Best Practices for Kubernetes-Based Workflow Orchestration****

To fully leverage Kubernetes’ power for workflow orchestration, following best practices that ensure scalability, resilience, and efficiency is crucial. Based on my experience designing and optimizing workflow systems at scale, here are the key best practices:

****Prioritize Stateless Architectures for Scalability****

Stateless architectures scale effortlessly because they don’t maintain an internal state between executions. This design is ideal for cloud-native environments where workloads can dynamically scale without persistent data storage. Stateless applications can be scaled horizontally by adding or removing container instances without affecting functionality.

In a cloud-native workflow I developed, we used stateless microservices for API processing. This allowed the application to scale efficiently, handling high-traffic periods while maintaining consistent performance.

****Use Kubernetes Operators for Workflow Automation****

Kubernetes Operators and Custom Resource Definitions (CRDs) automate complex workflows, encapsulating operational knowledge within Kubernetes. Operators simplify the deployment and management of systems like database clusters, machine learning pipelines, and distributed data processing.

In one of my Kubernetes-based projects, we implemented a custom Operator to streamline the deployment of multi-step data processing workflows: this improved consistency, reduced manual configuration, and enhanced system reliability.

****Implement Adaptive Resource Management with MAPE-K****

Adaptive resource management optimizes cloud infrastructure. Kubernetes achieves this with the MAPE-K Model—Monitor, Analyze, Plan, Execute, and Knowledge—which adjusts resources based on real-time demand.

In a cloud-native project, we implemented adaptive scaling to optimize costs and performance. A notable example is Flyte, where adaptive resource management using Kubernetes supported scalable workflow management for Freenome’s cancer detection research.

****Monitor and Optimize Continuously with Prometheus and Grafana****

Continuous monitoring ensures system health and performance. Prometheus and Grafana are popular tools for real-time monitoring and visualization. By monitoring key metrics like CPU, memory, and network usage, we can proactively identify and resolve issues before they impact workflow execution.

In one project, we used Prometheus to collect real-time metrics and set up Grafana dashboards for insights, allowing us to identify performance anomalies and optimize resource allocation.

Kubernetes integrates seamlessly with Continuous Integration and Continuous Deployment (CI/CD) pipelines, enabling automated code deployment, testing, and updates. This ensures rapid, consistent updates without manual intervention.

In a cloud-native project, we integrated Kubernetes with a CI/CD pipeline using Jenkins and GitLab CI, enabling automated deployments with zero downtime.

****Design for High Availability with Worker Pool Models****

The Worker Pool Model dynamically scales worker nodes based on demand, ensuring workflows run efficiently. This approach maximizes resource efficiency and availability, making it ideal for data-intensive or resource-heavy workflows.

Using this model, I could dynamically scale a distributed data processing system, optimizing performance and cost.

****Why Mastering Workflow Orchestration is Essential****

Workflow orchestration is vital to building scalable cloud infrastructure, and Kubernetes is the perfect platform. With my extensive experience in designing cloud-native systems, I’ve witnessed firsthand how well-executed workflow orchestration transforms cloud performance, enabling organizations to unlock the full potential of their infrastructure.

As cloud technology evolves, workflow orchestration will be at the heart of innovation. For anyone building scalable systems, mastering Kubernetes-based orchestration is not just a choice—it’s essential. Ready to take control of your cloud infrastructure and optimize your workflows? Let’s start a conversation.

****References:****

**Shan, C., et al. (2023)**. _An Efficient Data-Driven Workflow Automation Model for Scalable Cloud Systems._ arXiv. https://arxiv.org/abs/2301.08409 

**Flyte, (2023).** _Flyte’s Kubernetes-native Workflow Engine Propels Freenome’s Cancer Detection Research._ https://flyte.org/case-study/flytes-kubernetes-native-workflow-engine-propels-freenomes-cancer-detection-research 

**Orzechowski, M., Balis, B., Janecki, K., (2024).** _A Scalable Approach to Automating Complex Cloud Workflows using Kubernetes._ arXiv. https://arxiv.org/abs/2408.15445 

**Sengupta, S. (2022)**. _An Overview of CI/CD Pipelines with Kubernetes._ DZone. https://dzone.com/articles/an-overview-of-cicd-pipelines-with-kubernetes 

**Shan, C., et al. (2022).** _Kubernetes-Based Workflow Orchestration for Cloud-Native Systems._ arXiv. https://arxiv.org/abs/2207.01222

(Top, Featured Image Photo via Shutterstock)

From Complexity to Clarity: The Blueprint for Scalable Workflow Automation

We're rolling out a brand new feature in Malwarebytes for iOS: the ability to block Google sponsored ads directly on Safari.

Sponsored ads on Google search don’t just irritate users—they also provide a dangerous opportunity for cybercriminals to spread malware and scams to their unsuspecting victims. What looks like a harmless search result can be a carefully disguised trap.  

At Malwarebytes, our researchers have uncovered a variety of threats hiding in plain sight within these sponsored ads, including Mac stealers distributed through Google Ads, scams targeting popular utility software, and tech support traps.  

In some cases, scammers use advanced AI tools like DeepSeek AI to create convincing ads and web pages, increasing the risk to unsuspecting users.  

That’s why we’re excited to roll out a brand new feature in Malwarebytes for iOS: the ability to block Google sponsored ads directly on Safari.  

If you’ve used our iOS app before, you already know it blocks ads on webpages and ad trackers. But with our newest feature, we’re extending protection to cover those annoying and potentially dangerous sponsored ads listed on search results.  

Now, with just a simple toggle, you can make those annoying sponsored ads disappear from your Safari browsing experience.  

**Try it for yourself**: Download Malwarebytes for iOS and enjoy this feature—alongside scam text filtering, a privacy VPN, and call protection—with a free seven-day trial.  

Existing users should already have this feature option in their app—check it out now!

**Don’t have an iOS device?** Download our free Browser Guard extension for ad blocking and scam protection while browsing on your desktop.

 Tired of Google sponsored ads? So are we! That’s why we’re introducing the option to block them on iOS     

Bitdefender exposes Facebook ad scams using fake crypto sites and celebrity lures to spread malware via malicious desktop…

Bitdefender exposes Facebook ad scams using fake crypto sites and celebrity lures to spread malware via malicious desktop clients and PowerShell scripts.

A persistent malware campaign is exploiting Facebook’s advertising network to target cryptocurrency enthusiasts, security researchers at Bitdefender revealed today.

The operation leverages the trusted names of major cryptocurrency exchanges like Binance and TradingView, and images of celebrities such as Elon Musk and Zendaya in Facebook ads to lend credibility to the fake cryptocurrency exchange promotions and lure unsuspecting users into downloading malicious software.

Source: Bitdefender

Bitdefender’s investigation, shared with Hackread.com ahead of its publishing, found a multi-layered attack that delivers malware through a covert communication channel between the website and the victim’s own computer.

According to researchers, cybercriminals are hijacking Facebook accounts or creating fake ones to run deceptive ads promising quick financial gains or crypto bonuses. Clicking these ads redirects victims to convincing but fraudulent websites that mimic legitimate cryptocurrency platforms, urging them to download a “desktop client.”

When downloaded, the desktop client drops a malicious DLL file, which launches a local .NET\-based server on the victim’s machine. This server acts as a hidden C2 centre. The fake website’s front end contains a deobfuscated script that communicates with the server, sends WMI (Windows Management Instrumentation) queries, and instructs it to execute further malicious payloads.

The final stage often involves the execution of multiple encoded PowerShell scripts, which download additional malware from remote servers. Furthermore, the attackers implement advanced anti-sandbox checks, ensuring that the malware is only delivered to users who meet specific demographic and behavioural profiles deemed valuable by the cybercriminals.

Bitdefender researcher Ionut Baltariu highlighted that users without specific Facebook ad tracking parameters, those not logged into Facebook, or those with uninteresting IP addresses or operating systems are also shown harmless content instead. This targeted approach allows the attackers to maximize their impact while minimizing exposure to security analysis.

****100 Malicious Ads in Just 24 Hours****

The scale of the operation is significant as researchers have identified hundreds of Facebook accounts actively promoting these malicious pages. In one case, a single page ran over 100 ads in just 24 hours.

While Facebook often removes these fraudulent ads, many gather thousands of views before being taken down. The targeting is often finely tuned, with one instance focusing on men aged 18 and over in Bulgaria and Slovakia.

Adding another layer of deception, the attackers have even created fake Facebook pages that perfectly mirror the official pages of platforms like TradingView, complete with fabricated posts and comments touting fake giveaways. However, the links embedded in these fake pages lead directly to the malware-distributing websites.

Facebook’s continued role as a vector for malware distribution is hard to overlook as earlier findings, including today’s discovery from Morphisec which shows cybercriminals have been using deceptive Facebook ads promoting fake AI platforms to distribute the new Noodlophile Stealer.

It also shows how cybercriminals exploit the platform’s reach and advertising capabilities for malicious purposes, emphasizing the need for user vigilance and platform security enhancements.

Bitdefender advises users to be cautious of online ads, use scam and link-checking tools, keep security software updated and report suspicious ads on Facebook to stay protected.

Fake Crypto Exchange Ads on Facebook Spread Malware

The age of AI guessing our passwords is upon us, and we need to change the ways we authenticate and use passwords where we have no alternatives.

For decades, passwords have been our default method for keeping online accounts safe. But in the age of artificial intelligence, this traditional security method is facing challenges it was never built to withstand.

A team at Cybernews conducted a study of over 19 billion newly exposed passwords which showed we’re looking at a “a widespread epidemic of weak password reuse.” It shows that despite years of trying to educate users about the dangers of using weak, lazy passwords, and re-using them across different sites and services, we have hardly made any progress.

But our opponents have. They can use new tools, faster computers, and because of both these developments, they ended up needing less effort for a greater yield. Because our digital presence in life has grown enormously and with that the number of passwords and the importance of the information they can unlock.

**Enter AI**

Artificial Intelligence (AI)-powered tools are now capable of cracking passwords faster and more efficiently than ever before. What once took days or weeks using brute force can now be accomplished in minutes. Tools like PassGAN (Password Generative Adversarial Network) use deep learning to predict and generate likely passwords based on leaked data sets. Unlike traditional dictionary attacks, AI doesn’t rely solely on existing word lists. AI is able to learn patterns from billions of compromised passwords and create new ones that closely mimic real human behavior.

This represents a huge advantage to the attackers. While a human hacker might guess that someone used their pet’s name followed by the year they were born, an AI can deduce that “Fluffy2023!” is statistically probable based on thousands of other similar combinations. And it can do this millions of times per second.

AI’s password-cracking capabilities are further supercharged by powerful hardware. Graphics processing units (GPUs), which are commonly used in gaming and scientific computing, can now be harnessed to run password-cracking algorithms at scale. Combined with AI, these machines make short work of weak or even moderately complex passwords.

The result is a world where even passwords once considered strong, like for example “Tr33House!” may no longer provide meaningful protection.

**Does that make the password obsolete?**

Tech companies are already betting on a passwordless future. Passkeys, biometrics, and multi-factor authentication (MFA) are gaining traction. Passkeys, in particular, offer a cryptographic alternative that eliminates the need for users to remember or even create passwords at all. But adoption of passkeys is still in the early stages, and many systems still rely on traditional passwords.

Beyond the technical risks, there are serious personal consequences when passwords are stolen. Due to our widespread online presence, once an attacker obtains your login credentials, they can access sensitive documents, reset other account passwords, or impersonate you online. From there, the path to identity theft is short. Criminals can use stolen data to open credit lines, file fraudulent tax returns, or drain your savings. In many cases, victims don’t even know their identity has been stolen until serious financial or legal damage has already occurred.

In the age of AI, the stakes are higher, and the window of vulnerability is shorter. A single reused or weak password might be all it takes to lose control over your digital identity.

The lesson is clear: we can’t rely on passwords alone anymore. AI has changed the game even further, and now it’s up to us to change how we play it. And as far as passwords go, there are some ways to use them as securely as possible where you have no alternative:

*   Make passwords as strong as possible and never reuse passwords.
*   Use a password manager to help remember all the passwords.
*   Where possible, use MFA as an extra layer.
*   Pressure important services into adapting passkeys and use them as soon as the occasion arises.

You can use Malwarebytes’ free Digital Footprint scan to see how many passwords of yours have been included in leaks and data breaches.

**We don’t just report on threats – we help safeguard your entire digital identity**

Cybersecurity risks should never spread beyond a headline. Protect your—and your family’s—personal information by using identity protection.

 Passwords in the age of AI: We need to find alternatives 

Scammers are using fake AI tools and Facebook ads to spread Noodlophile Stealer malware, targeting users with a…

Scammers are using fake AI tools and Facebook ads to spread Noodlophile Stealer malware, targeting users with a multi-stage attack to steal credentials.

Cybersecurity researchers at Morphisec have identified a new malware campaign utilizing fake and malicious artificial intelligence (AI) platforms to distribute a new information stealer dubbed Noodlophile Stealer.

This sophisticated tactic exploits the surging popularity of AI tools to trick users into downloading malware that can steal browser credentials, and cryptocurrency wallets, and potentially deploy remote access tools like XWorm.

****How Does it Work?****

Morphisec’s threat analysis, shared with Hackread.com ahead of its publishing on 8 May 2025, details how cybercriminals are creating convincing fake AI websites, often advertised through Facebook groups with global reach (some posts exceeding 62,000 views on a single post).

These platforms lure users with promises of free AI video and image generation, prompting them to upload their own images. Instead of the expected AI-processed content, victims unknowingly download a malicious ZIP archive containing the Noodlophile Stealer.

Source: Morphisec

This campaign stands out due to its exploitation of AI as a social engineering lure, targeting a potentially more trusting audience of creators and small businesses exploring AI, mainly within Facebook communities.

Morphisec report notes that Noodlophile Stealer is a newly documented malware combining credential theft, wallet exfiltration, and optional remote access deployment. Notably, it exfiltrates stolen information through a Telegram bot.

Open-source intelligence (OSINT) investigations led Morphisec to identify the developer behind Noodlophile, likely of Vietnamese origin, who was observed promoting this method in Facebook posts and on online cybercrime marketplaces. The developer’s profile also reveals further involvement in malware sales and distribution, with links found in Facebook groups leading directly to their profile.

Source: Morphisec

****Multi-Stage Attack Designed for Evasion****

The attack chain involves a multi-stage infection process designed for stealth and persistence. Users interacting with the fake AI site download a ZIP file (VideoDreamAI.zip) containing a deceptive executable (Video Dream MachineAI.mp4.exe), which is a repurposed version 445.0 of the legitimate video editing tool, CapCut, and is even signed using a certificate created via Winauth.

This executable then drops further malicious components from a hidden folder named 5.0.0.1886, including CapCut.exe (a wrapper for embedded .NET malware), AICore.dll (a command execution helper), and disguised files like Document.docx (a batch script) and Document.pdf (a password-protected archive).

The install.bat script, launched by CapCutLoader (within CapCut.exe, which first verifies internet connectivity by pinging google.com up to 10 times), decodes the archive (password: TONGDUCKIEMDEVELOPER2025), establishes persistence, and downloads and executes a Python payload (srchost.exe) containing the Noodlophile Stealer and the XWorm loader.

These final payloads operate in memory to evade detection, with the XWorm loader employing techniques like shellcode injection and PE hollowing (especially targeting RegAsm.exe if Avast is present).

The Noodlophile Stealer and its use of fake AI platforms is just another cybersecurity threat against unsuspected users. Therefore, one must remain cautious at all times, refrain from downloading tools directly from social media posts or third-party platforms, and always use official websites to download files.

Even after downloading a file from a verified source, do not execute/install the program on your device before scanning it on websites like VirusTotal or ANY.RUN.

Fake AI Tools Push New Noodlophile Stealer Through Facebook Ads

A new spam campaign is targeting Brazilian users with a clever twist — abusing the free trial period of trusted remote monitoring tools and the country’s electronic invoice system to spread malicious agents.

*   Cisco Talos identified a spam campaign targeting Brazilian users with commercial remote monitoring and management (RMM) tools since at least January 2025. Talos observed the use of PDQ Connect and N-able remote access tools in this campaign. 
*   The spam message uses the Brazilian electronic invoice system, NF-e, as a lure to entice users into clicking hyperlinks and accessing malicious content hosted in Dropbox. 
*   Talos has observed the threat actor abusing RMM tools in order to create and distribute malicious agents to victims. They then use the remote capabilities of these agents to download and install Screen Connect after the initial compromise.
*   Talos assesses with high confidence that the threat actor is an initial access broker (IAB) abusing the free trial periods of these RMM tools. 

Talos recently observed a spam campaign targeting Portuguese-speaking users in Brazil with the intention of installing commercial remote monitoring and management (RMM) tools. The initial infection occurs via specially crafted spam messages purporting to be from financial institutions or cell phone carriers with an overdue bill or electronic receipt of payment issued as an NF-e (see Figures 1 and 2). 

Figure 1. Spam message purporting to be from a cell phone provider. 

Figure 2. Spam message masquerading as a bill from a financial institution. 

Both messages link to a Dropbox file, which contains the malicious binary installer for the RMM tool. The file names also contain references to NF-e in their names: 

*   AGENT\_NFe\_<random>.exe 
*   Boleto\_NFe\_<random>.exe 
*   Eletronica\_NFe\_<random>.exe 
*   Nf-e<random>.exe 
*   NFE\_<random>.exe 
*   NOTA\_FISCAL\_NFe\_<random>.exe 

_Note: <random> means the filename uses a random sequence of letters and numbers in that position._ 

The victims targeted in this campaign are mostly C-level executives and financial and human resources accounts across several industries, including some educational and government institutions. This assessment is based on the most common recipients found in the messages Talos observed during this campaign. 

Figure 3. Targeted recipients.

This campaign's objective is to lure the victims into installing an RMM tool, which allows the threat actor to take complete control of the target machine. N-able RMM Remote Access is the most common tool distributed in this campaign and is developed by N-able, Inc., previously known as SolarWinds. N-able is aware of this abuse and took action to disable the affected trial accounts. Another tool Talos observed in some cases is PDQ Connect, a similar RMM application. Both provide a 15-day free trial period.

To assess whether these actors were using a trial version rather than stolen credentials to create these accounts, Talos checked samples older than 15 days and confirmed all of them returned errors that the accounts were disabled, while newer samples found in the last 15 days were all active.

Talos also examined the email accounts used to register for the service. They all use free email services such as Gmail or Proton Mail, as well as usernames following the theme of the spam campaign, with few exceptions where the threat actors used personal accounts. These exceptions are potentially compromised accounts which are being abused by the threat actors to create additional trial accounts. Talos did not find any samples in which the registered account was issued by a private company, so we can assess with high confidence these agents were created using trial accounts instead of stolen credentials.

_N-able is aware of this abuse and took action to disable the affected trial accounts._

Talos found no evidence of a common post-infection behavior for the affected machines, with most machines staying infected for days before any other malicious activity was executed by the tool. However, in some cases, we observed the threat actor installing an additional RMM tool and removing all security tools from the machine a few days after the initial compromise. This is consistent with actions of initial access broker (IAB) groups. 

An IAB's main objective is to rapidly create a network of compromised machines and then sell access to the network to third parties. Threat actors commonly use IABs when looking for specific target companies to deploy ransomware on. However, IABs have varied priorities and may sell their services to any threat actors, including state-sponsored actors.  

Adversaries’ abuse of commercial RMM tools has steadily increased in recent years. These tools are of interest to threat actors because they are usually digitally signed by recognized entities and are a fully featured backdoor. They also have little to no cost in software or infrastructure, as all of this is generally provided by the trial version application.  

Talos created a trial account to test what features were available for a trial user. In the case of the N-able remote access tool, the trial version offers a full set of features only limited by the 15-day trial period. Talos was able to confirm that by using a trial account, the threat actor has full access to the machine, including remote desktop like access, remote command execution, screen streaming, keystroke capture and remote shell access. 

Figure 4. N-able management interface showing available remote access tools. 

Figure 5. Administrative shell executed on a remote machine. 

The threat actor also has access to a fully featured file manager to easily read and write files to the remote file system. 

Figure 6. N-able file manager. 

The network traffic these tools create is also disguised as regular traffic, with many tools using communication over HTTPS and connecting to resources which are part of the infrastructure provided by the application provider. For example, N-able Remote Access uses a domain associated with its management interface, hosted on Amazon Web Services (AWS): 

*   hxxps://upload1\[.\]am\[.\]remote\[.\]management/ 
*   hxxps://upload2\[.\]am\[.\]remote\[.\]management/ 
*   hxxps://upload3\[.\]am\[.\]remote\[.\]management/ 
*   hxxps://upload4\[.\]am\[.\]remote\[.\]management/ 

_**Disclaimer**: The URLs above are part of the management infrastructure for the RMM tools described in this blog and are not controlled by the threat actor. Customers must complete an assessment before enabling block signatures for these domains._ 

The domain the agent uses is the same for any customer using the tool, with only the username and API key differentiating which customer the agent belongs to, as can be seen in Figure 7. This makes it even more difficult to identify the origin of the attacks and perform threat actor attribution.

Figure 7. Example configuration file. 

By extracting the configuration files inside the agent installer files still available on Dropbox, we can see some email addresses follow the same theme of the spam emails, using names of finance-related users and domains, while others could be potentially compromised accounts being used to create trial accounts for N-able Remote Access.  

With these trial versions being limited only by time and providing full remote-control features with little to no cost to the threat actors, Talos expects these tools to become even more common in attacks. 

Cisco Secure Firewall Application control is able to detect the unintended usage of RMM tools in customer's networks. Instructions on how to set up Application control can be found at Cisco Secure Firewall documentation. 

**Coverage **

Ways our customers can detect and block this threat are listed below. 

Cisco Secure Endpoint (formerly AMP for Endpoints) is ideally suited to prevent the execution of the malware detailed in this post. Try Secure Endpoint for free here. 

Cisco Secure Email (formerly Cisco Email Security) can block malicious emails sent by threat actors as part of their campaign. You can try Secure Email for free here. 

Cisco Secure Firewall (formerly Next-Generation Firewall and Firepower NGFW) appliances such as Threat Defense Virtual, Adaptive Security Appliance and Meraki MX can detect malicious activity associated with this threat. 

Cisco Secure Network/Cloud Analytics (Stealthwatch/Stealthwatch Cloud) analyzes network traffic automatically and alerts users of potentially unwanted activity on every connected device. 

Cisco Secure Malware Analytics (Threat Grid) identifies malicious binaries and builds protection into all Cisco Secure products. 

Cisco Secure Access is a modern cloud-delivered Security Service Edge (SSE) built on Zero Trust principles.  Secure Access provides seamless transparent and secure access to the internet, cloud services or private application no matter where your users work.  Please contact your Cisco account representative or authorized partner if you are interested in a free trial of Cisco Secure Access. 

Umbrella, Cisco’s secure internet gateway (SIG), blocks users from connecting to malicious domains, IPs and URLs, whether users are on or off the corporate network.  

Cisco Secure Web Appliance (formerly Web Security Appliance) automatically blocks potentially dangerous sites and tests suspicious sites before users access them.  

Additional protections with context to your specific environment and threat data are available from the Firewall Management Center. 

Cisco Duo provides multi-factor authentication for users to ensure only those authorized are accessing your network.  

Open-source Snort Subscriber Rule Set customers can stay up to date by downloading the latest rule pack available for purchase on Snort.org. 

ClamAV detections are also available for this threat:  

Txt.Backdoor.NableRemoteAccessConfig-10044370-0  
Txt.Backdoor.NableRemoteAccessConfig-10044371-0   
Txt.Backdoor.NableRemoteAccessConfig-10044372-0 

**Indicators of Compromise **

_**Disclaimer**: The URLs below are part of the management infrastructure for the RMM tools described in this blog and are not controlled by the threat actor. An assessment must be done by customers before enabling block signatures for these domains._ 

_IOCs for this threat can be found on our GitHub repository_ _here__._ 

**Network IOCs **

hxxps://upload1\[.\]am\[.\]remote\[.\]management/   
hxxps://upload2\[.\]am\[.\]remote\[.\]management/   
hxxps://upload3\[.\]am\[.\]remote\[.\]management/   
hxxps://upload4\[.\]am\[.\]remote\[.\]management/   
198\[.\]45\[.\]54\[.\]34\[.\]bc\[.\]googleusercontent\[.\]com 

**RMM Installer - Hashes **

03b5c76ad07987cfa3236eae5f8a5d42cef228dda22b392c40236872b512684e   
0759b628512b4eaabc6c3118012dd29f880e77d2af2feca01127a6fcf2fbbf10   
080e29e52a87d0e0e39eca5591d7185ff024367ddaded3e3fd26d3dbdb096a39   
0de612ea433676f12731da515cb16df0f98817b45b5ebc9bbf121d0b9e59c412   
1182b8e97daf59ad5abd1cb4b514436249dd4d36b4f3589b939d053f1de8fe23   
14c1cb13ffc67b222b42095a2e9ec9476f101e3a57246a1c33912d8fe3297878   
2850a346ecb7aebee3320ed7160f21a744e38f2d1a76c54f44c892ffc5c4ab77   
4787df4eea91d9ceb9e25d9eb7373d79a0df4a5320411d7435f9a6621da2fd6b   
51fa1d7b95831a6263bf260df8044f77812c68a9b720dad7379ae96200b065dd   
527a40f5f73aeb663c7186db6e8236eec6f61fa04923cde560ebcd107911c9ff   
57a90105ad2023b76e357cf42ba01c5ca696d80a82f87b54aea58c4e0db8d683   
63cde9758f9209f15ee4068b11419fead501731b12777169d89ebb34063467ea   
79b041cedef44253fdda8a66b54bdd450605f01bbb77ea87da31450a9b4d2b63   
a2c17f5c7acb05af81d4554e5080f5ed40b10e3988e96b4d05c4ee3e6237c31a   
b53f9c2802a0846fc805c03798b36391c444ab5ea88dc2b36bffc908edc1f589   
c484d3394b32e3c7544414774c717ebc0ce4d04ca75a00e93f4fb04b9b48ecef   
ca11eb7b9341b88da855a536b0741ed3155e80fc1ab60d89600b58a4b80d63a5   
d1efebcca578357ea7af582d3860fa6c357d203e483e6be3d6f9592265f3b41c   
e2171735f02f212c90856e9259ff7abc699c3efb55eeb5b61e72e92bea96f99c   
e34b8c9798b92f6a0e2ca9853adce299b1bf425dedb29f1266254ac3a15c87cd   
ebdefa6f88e459555844d3d9c13a4d7908c272128f65a12df4fb82f1aeab139f   
f52b4d81c73520fd25a2cc9c6e0e364b57396e0bb782187caf7c1e49693bebbf   
f5efd939372f869750e6f929026b7b5d046c5dad2f6bd703ff1b2089738b4d9c   
F68ae2c1d42d1b95e3829f08a516fb1695f75679fcfe0046e3e14890460191cf   
a71e274fc3086de4c22e68ed1a58567ab63790cc47cd2e04367e843408b9a065

Spam campaign targeting Brazil abuses Remote Monitoring and Management tools

In the wake of SignalGate, a knockoff version of Signal used by a high-ranking member of the Trump administration was hacked. Today on Uncanny Valley, we discuss the platforms used for government communications.

All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links.

When former national security adviser Mike Waltz had a picture taken of him last week, he didn’t expect for the whole world to see that he was using TeleMessage, a messaging app similar to Signal. Now the app has been hacked, with portions of data linked to government entities like Customs and Border Protection (CBP) and companies like Coinbase. Today on the show, we’re joined by WIRED senior writer Lily Hay Newman to discuss what this incident tells us about the growing vulnerabilities in government communications.

Articles mentioned in this episode:  
**Mike Waltz Has Somehow Gotten Even Worse at Using Signal**, by Lily Hay Newman  
**The Signal Clone the Trump Admin Uses Was Hacked** , by Joseph Cox and Micah Lee  
**The Signal Clone Mike Waltz Was Caught Using Has Direct Access to User Chats**, by Lily Hay Newman

You can follow Zoë Schiffer on Bluesky at @zoeschiffer and Lily Hay Newman on Bluesky at @lhn. Write to us at uncannyvalley@wired.com.

**How to Listen**

You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here’s how:

If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “Uncanny Valley.” We’re on Spotify too.

**Transcript**

_Note: This is an automated transcript, which may contain errors._

**Zoë Schiffer:** Hi, this is Zoë. Before we start, I want to take the chance to remind you that we want to hear from you. If you have tech-related questions that have been on your mind or a topic that you wish we'd cover, write to us at uncannyvalley@WIRED.com. And if you listen to and enjoy the show, please rate it and leave a review on your podcast app of choice. It really honestly makes a difference. Welcome to WIRED's _Uncanny Valley_. I'm WIRED's director of business and industry, Zoë Schiffer. Today on the show, the hacking scandal surrounding TeleMessage, the knockoff version of Signal, which is used by at least one high-ranking member of the Trump administration. The app has temporarily suspended its services while it investigates the incident. We're going to talk about how former national security adviser Mike Waltz was seen last week using the app in a cabinet meeting and what this latest incident tells us about the growing vulnerabilities in government communication. I'm joined by Lily Hay Newman, senior writer at WIRED. Lily, welcome to the show.

**Lily Hay Newman:** It's a pleasure to be here.

**Zoë Schiffer:** What exactly is TeleMessage?

**Lily Hay Newman:** Yeah. So TeleMessage is a company that's been around since the late ’90s. It was founded in Israel, and it creates apps that are sort of mirror images or clones of existing communication apps, and then adds in an archiving feature. So this is especially perhaps wanted for apps that are securing communications, such that it's difficult to retain copies of the messages. So if you need copies for compliance or you need a record, the idea is that these services are giving the same functionality as apps you know, like WhatsApp or Telegram or Signal, but with the addition of these archiving features.

**Zoë Schiffer:** And that's important, obviously, for people who work in government because, technically, members of the press and other people are supposed to be allowed to access a lot of the communications that aren't classified by submitting Freedom of Information Act requests. And you can't do that if the messages are disappearing.

**Lily Hay Newman:** Correct. There are record retention laws in the US and other countries for transparency and information requests, as you said. But historically, the way governments and other institutions have complied with that is by using communication platforms that are built for the purpose of government communications, tailor-built to be in compliance in a number of ways. So all of this is coming up because now the Trump administration in recent months has been sort of departing from the standard ways that officials in the US have communicated to use consumer platforms, particularly the secure messaging platform Signal, to talk to each other, but doing so in a very ad hoc consumer way like in the same way that you and I would set up a Signal conversation. That's what they've been doing, and that's where you get into this whole question of how do you comply with records requirements. How do you comply with safety requirements when you're just kind of using off-the-shelf tech in a regular way? And so that's where TeleMessage comes in.

**Zoë Schiffer:** Well, it seems like one of the people, as we mentioned earlier, who was using TeleMessage was Mike Waltz, the now former national security adviser, who at this point is best known for starting that infamous Signal group chat a few weeks back that accidentally added a senior member of The Atlantic Newsroom. How did we find out that he was using TeleMessage in the first place?

**Lily Hay Newman:** So his screen, the screen of his phone, was sort of inadvertently captured in a photo of a cabinet meeting, a Reuters photo, that Mike Waltz was participating in, was sitting at the table with Trump and a number of officials. The photo is a bit funny because it seems like he thinks no one can see him using his phone, or he is kind of checking his phone. I mean, we've all been there, looking under the conference table at our phone. But additionally, his screen shows what appears to be Signal. So we're really going, zooming in deep into this photo, right. We're looking over his shoulder at his phone. Now we're seeing this notification. And then in the notification, instead of the normal words that would be there, people noticed that the Signal … where it would normally say Signal, was being referred to as TM Signal. And that's how people realized that, actually, he was using this other app called TeleMessage.

**Zoë Schiffer:** Got it. Yeah. Nothing makes me love reporters more than the absolute psychotic behavior of zooming in on a tiny little phone screen to be like, “What exactly is going on here?” But kudos to 404 Media, because I think they were the first ones to point that out. You wrote in a recent WIRED article that Mike Waltz has inexplicably gotten even worse at using Signal. So, I guess what did you mean by that? How is he getting worse at using this end-to-end encrypted app?

**Lily Hay Newman:** This whole revelation about his use of TM Signal is building on this previous situation called Signal Gate. Mike Waltz was the person who inadvertently added Jeffrey Goldberg, the top editor of The Atlantic, to the chat. And so already Mike Waltz was not having a great track record, and then disappearing messages were on the whole time. And so, one of the many criticisms was that this was not in compliance with government record-retention laws. So we don't know this, but presumably then he started using TM Signal as a solution to that aspect of the issues raised. But I just want to be clear. We don't know. It could be that they were already using it, or he was already using TM Signal at the time. I'm not sure. But one might suspect that hearing some of this criticism, he was like, “OK, let me find a solution that does retain records and does have an archiving feature.” And that's where TeleMessage would come in.

**Zoë Schiffer:** So the national security advisoer sets up this group chat, presumably not in compliance, then switches to one that looks like it might be in compliance, and then that version is promptly hacked. Do we know at this point who is behind the hacking?

**Lily Hay Newman:** More and more is coming out about potential hacks of TeleMessage or sort of ability to intercept messages and see messages in memory. First, 404 Media and Micah Lee published a piece with an unnamed hacker providing evidence that they could breach TeleMessage. And then, on Monday, NBC News published an additional report with an additional unnamed hacker. So clearly there's a lot of insecurity here. And the criticism of TM Signal from this company, TeleMessage, is that it claims to have all the same security features as real Signal and to sort of preserve that, and just add on this archiving feature. But, definitionally, adding in the archiving feature breaks Signal security. The way signal is designed and other end-to-end encrypted apps like WhatsApp, when you add in this other party, it's virtually impossible that the security guarantees could be preserved. And then, on top of that, it seems like from source code review that's starting to come out, and research that's starting to happen, and analysis into TM Signal, that actually it's just not constructed in a very secure way at all. So, just a lot of layers to get to the point, which is that this was a wildly insecure app for Mike Waltz to be using, sitting at a table with the top cabinet members and the president of the United States. It's wild.

**Zoë Schiffer:** We're going to get into what exactly was accessed in this hack. But before we do that, we're going to take a short break.

\[_break_\]

**Zoë Schiffer:** We are back. So let's get into what exactly was accessed when it looks like multiple hackers were able to break into TM Signal, which was being used by at least one member of the Trump administration.

**Lily Hay Newman:** So far, these researchers, what they've shown is that some messages, sometimes at least, are being sent to the archiving server in plain text, meaning they are readable. That's precisely what a platform like genuine Signal is trying to avoid. And so that's what's happening. So these were sort of fragments or pieces or whole messages, but not whole conversations, things like that, so far. One thing that 404 Media reported on from these leaks was evidence that US Customs and Border Patrol agents have been using TM Signal. It's not totally clear what's going on with this. WIRED reached out to CBP. We've been trying to get clarification on what this leaked data means. There seem to be confirmed CBP phone numbers associated with these accounts that came out of this breach. CBP has told WIRED just that they're looking into it. But that's an example that is really concerning, it would potentially show that this app is in wider use across other agencies in the US government.

**Zoë Schiffer:** Is there a national security concern with the fact that this app was developed in Israel, regardless of the fact that it was acquired by a US company recently?

**Lily Hay Newman:** The thing is, even without getting into any specific geopolitics, the point of the protocols that exist for the US government to use its own purpose-built communication platforms is that any and all foreign governments conduct espionage. The US does it. Everyone does it. So, for your most sacred and sensitive national communication, you want to do that on a platform that you completely control, that you have built and vetted yourself, and just all parameters are controlled by you. You don't want to involve any other parties. So Israeli espionage groups are known for being very aggressive, very innovative, very cunning. So, for that reason, particularly, perhaps it's a concern that TeleMessage was founded in the country and has those ties. But just in general, regardless of what country it is, I think it's important conceptually to understand that it doesn't make sense to use the app in this way.

**Zoë Schiffer:** After this reporting came out, TeleMessage has paused or stopped its services. What's the status of the company right now?

**Lily Hay Newman:** Right. So clearly, they have concerns, and their parent company, Smarsh, has concerns about these findings as well. They say that they are investigating a potential breach and have employed a third-party firm to help them with that. And they've taken down all the content from the TeleMessage website and paused TeleMessage operations, essentially. So they say it's a pause and pending the investigation, but a pretty big reaction here to these findings.

**Zoë Schiffer:** That's a good place to end it. When we come back, we'll share our recommendations for what to check out on WIRED.com this week. Welcome back to _Uncanny Valley_. I'm Zoë Schiffer, WIRED's director of business and industry. I'm joined today by WIRED senior writer Lily Hay Newman. Before we take off, Lily, tell our listeners what they absolutely have to read on WIRED this week.

**Lily Hay Newman:** I'm just fascinated by this story by our colleague Caroline Haskins. US border agents are asking for help taking photos of everyone entering the country by car. And this is, we're just continuing our CBP discussions for today. CBP has apparently released a request for information seeking pitches, essentially for companies to help them do vehicle surveillance at the border and face recognition technology to see specifically who is in cars, not just the front seat. And I think it's really important for all of us to be aware of the extensive and expansive surveillance dragnet at the US border and all different types of US border crossings. The southern border of the US has long been known as sort of like a forefront of surveillance technology. And so it's dark, but interesting to hear that CBP feels like they don't yet have what they need to do this type of analysis and face recognition in cars, but that they want it, and they're trying to expand the analysis they can do on who is in every car.

**Zoë Schiffer:** Right. And it'll be interesting to see which company gets this contract. OK. Well, I wanted to flag a piece that we published yesterday by Paresh Dave and Kylie Robison. It's about OpenAI announcing that it is not, in fact, going to restructure its company to make the nonprofit arm not in control. In other words, the nonprofit arm is going to remain in control of the company. And this is a reversal of a prior announcement where it said it was going to become a public benefit corporation, likely to make fundraising easier. But after the plan was announced, the company got a ton of pushback from a variety of civic organizations and also Elon Musk, who was involved in the founding of the company before an acrimonious split in 2018. These groups don't usually agree on a lot, but they agreed on this, that becoming a for-profit company was in violation of OpenAI's founding mission. So we have a lot of good reporting on how people are taking this news and what it means for the future of the company. That's our show for today. We'll link to all the stories we spoke about in the show notes. Make sure to check out Thursday's episode of _Uncanny Valley_, which is about Trump's meme coin saga and the conflict of interest that come with it. Adriana Tapia produced this episode. Amar Lal at Macro Sound mixed this episode. Jordan Bell is our executive producer. Condé Nast's head of global audio is Chris Bannon. And Katie Drummond is WIRED's global editorial director.

Tag

#mac