Black Friday is an annual bargain hunt that often spirals into chaotic impulse buying. Stores promise incredible discounts of 50–70%, but are those savings really as significant as they seem? In 2025, we’ve got a new ally on our side in the fight for smart spending: artificial intelligence. Here’s how you can use powerful LLMs like ChatGPT and Claude to save money and never fall for a shady seller’s tricks again.

Before we enlist AI to help you save, it’s crucial we understand the battlefield. Studies paint a grim picture: a significant portion of those Black Friday “super discounts” are nothing more than a marketing illusion.

The tactic is simple and effective: in early October, stores hike up their prices, sometimes by fifty to a hundred percent. Then, when Black Friday finally hits, they “slash” the price by that same 50% and proudly tout the impressive discount on the tag. In reality, you’re just buying the item at its regular price — or sometimes even paying a premium.

While the European Union’s Omnibus Directive mandates that retailers display the lowest price from the last 30 days, even this rule is easily skirted. Retailers just hike the price up 30 days before the event, which allows them to technically adhere to the directive while still duping consumers.

How LLMs can help you save

Artificial intelligence is changing the game. Analysts estimate that in 2024, AI tools helped consumers make a staggering $60 billion in transactions during Cyber Week, and that number is only projected to climb in 2025. Already, one in three U.S. shoppers plans to lean on AI for their shopping needs.

As you know, an LLM is immune to emotion; it won’t react to marketing triggers like “2 hours left!” or “only one left in stock!” Instead, the model analyzes huge volumes of data, compares prices, tracks price history, and helps you make rational decisions.

In seconds, AI can crawl hundreds of online stores, zeroing in not only on the product you want at the lowest price but also on cheaper alternatives with comparable specs. Modern LLMs can help you figure out if a discount is truly beneficial — or if you’re falling for a scam. Amazon, for example, has already integrated a price-tracking feature into its AI assistant, Rufus, though users have noted that the tool still has some kinks to work out. Using just a few prompts, the AI can factor in your preferences, budget, and past purchases to suggest exactly what you need, cutting through all the marketing noise. Instead of wasting hours poring over spec sheets, just ask the assistant, “What’s the difference between vacuum cleaner A and vacuum cleaner B?” And you get your answer — regardless of whether the seller’s website features a comparison tool. You can use the prompts below for ChatGPT, Claude, or Gemini.

Preparing for Black Friday with AI

Step 1. Create a wish list

Don’t wait for the sales to start; your goal is to gather all the baseline data upfront.

Help me create a shopping list for Black Friday. My budget is: [amount].
I'm interested in the following categories: [electronics/clothing/home goods].
Priorities: [performance/quality/brand/price].
Create a structured list with explanations of why each item is worth considering.

Step 2. Start tracking prices

This is a critical stage. You need to know the real price of an item before the Black Friday marketing hype machine starts rolling. On Amazon, tools like CamelCamelCamel and Keepa can help, and for AliExpress, look at AliPrice and AliTools.

Step 3. Analyze price dynamics

Collected the price data? Excellent. If you see a sharp price spike in October followed by a corresponding drop in November, you’re looking at the classic scam tactic. But if the data on the charts seems unclear, use the prompt below. The months we used are just examples, so feel free to use your own date ranges. The larger the intervals between the price checks, the higher your chances of catching an unjustified price hike.

I'm tracking [product name] on [platform]. Here's the price data:
- September: [price]
- Early October: [price]
- Late October: [price]
- Current price: [price]
- Advertised discount: [percentage]
- Analyze this data. Is this a genuine discount or is the store manipulating prices?
When is the best time to buy? Should I wait for Black Friday or buy now?

Step 4. Search for alternatives

Don’t get fixated on a single product. There may be more advantageous alternatives available.

I want to buy [product, model]. My goal is to [what it's needed for]. Budget: [amount].
Find 3–5 alternative products that solve the same problem but might be more cost-effective.
Compare them based on features, price, and reviews. Display the results in a table.

Experience shows that LLM models are particularly good at comparative analysis, highlighting key differences between similar products.

Step 5. Vet the seller and the website

Black Friday is an absolute field day for scammers. In the third quarter of 2025 we saw the number of fake online stores skyrocket by 20% compared to the monthly average. Let’s run through the immediate red flags that should raise your suspicions:

• Domains like .shop, .store, .vip or .top — rarely used by major, established brands
• Unbelievable discounts of 80–90% on popular items
• Lack of a secure HTTPS connection, meaning no padlock icon next to the URL in your browser
• Poorly translated text and/or grammatical errors

Finally, just in case, run the following prompt through the AI of your choice to check the store’s legitimacy:

I have found [product name] on [URL]. The price is very attractive: [price], which is [percentage]% below the average. How can I verify that this is not a scam?
What are the signs of a fake store? What should I pay attention to?

Step 6. Compile the all-in-one prompt

This is the all-in-one prompt containing all the data you gathered in the previous steps; it works in any LLM:

You are an expert in spotting retail price manipulation.
Product: [name]
Store: [name]
Current price: [price]
Advertised discount: [percentage]%
Stated old price [price]
Price history I tracked:
[state data for several months]
Tasks:
1. Is this a genuine discount or a manipulation?
2. What was the real average price before the alleged sale?
3. Should I buy now, or is the price likely to drop even further?
4. Your verdict: buy / wait / look for alternatives?

Note that neural networks’ cybersecurity is still far from perfect: vulnerabilities continue to be discovered within them. Therefore, to shield yourself from phishing and spam links you might accidentally follow, be sure to install a proven and reliable security solution, such as Kaspersky Premium. It’ll keep your Black Friday from turning into a financial Black Monday for both your assets and personal data.

Getting local results

One of the core issues with global AI models is that they often deliver information that’s not region-specific, or is relevant to a region other than yours. But you can adapt them to your needs with this prompt:

You are an AI shopping assistant for [country, city]. All your recommendations must factor in the local market, available stores, and regional platforms ([list of stores, if desired]). State prices in [currency]. Speak [language].
My task is to find [product] at the best price for Black Friday.
Which local platforms should I check? What kind of sales are common in [region]?

Specialized prompts for each LLM

Each LLM has its strengths (also weaknesses). With these in mind, we’ve created prompts that unlock the potential of each language model. For the highest quality results, we recommend utilizing models with a larger number of parameters (usually available via paid subscriptions), and activating deep thinking when submitting your requests.

ChatGPT excels at structuring information and generating lists. Here’s a prompt for budget planning:

Create a shopping strategy for Black Friday.
Budget: [amount]
Priority categories: [list]
For each category, specify:
1. Average price before discounts
2. Expected discounted price
3. Best time to buy (before/during/after Black Friday)
4. Alternatives
Format the results as a table.

And here’s a prompt for store comparison:

Product: [name and model]
Found in stores:
- [Store 1]: [price], shipping [terms]
- [Store 2]: [price], shipping [terms]
- [Store 3]: [price], shipping [terms]
Which option is more cost-effective considering the total cost? Analyze the reliability of the stores.

Claude is particularly good at analyzing large volumes of text and highlighting key points. Here’s a Claude prompt for analyzing reviews:

Here's a selection of reviews for [name] from various platforms: [insert reviews].
Analyze them and highlight:
1. Key advantages (top 3)
2. Key disadvantages (top 3)
3. Who is this product best suited for, and who should avoid it?
4. Are there any alarming issues mentioned?
5. Overall recommendation: is this worth buying?

Long-term planning prompt:

You're a financial consultant. I'm planning a major purchase: [product] for [price].
My monthly income: [amount]. My savings: [amount].
Should I buy this on Black Friday or should I wait?
What alternative saving and purchasing strategies can you offer?

Gemini offers seamless integration with the Google ecosystem and provides in-depth capabilities when working with images. Attach a screenshot of the banner or the offer on the website and write the prompt:

This is a Black Friday offer. Evaluate:
1. How attractive is this discount?
2. What information should I check additionally?
3. What should I pay attention to in the description?
4. Signs of a possible scam

Quick search prompt:

Find the best Black Friday 2025 offers in [category].
I'm looking for: [product characteristics]
Budget: [amount]
Region: [country/city]
Show the top-5 options and provide a justification for each choice.

Final checklist

1. Use AI to create a wish list, and start tracking prices with tools like CamelCamelCamel, Keepa, or other similar services. Set up convenient price-drop notifications.
2. Analyze the collected price data, find alternative products and stores, and simultaneously verify the sellers’ reliability.
3. Set up a separate credit card for purchases with a spending limit. If possible, get a virtual card and prepare our prompts for quick retail-offer analysis.
4. On the actual sale day, don’t fall for urgency tricks like “last item in stock!”, and make sure you check every “super deal” with your AI assistant and a critical eye. Cross-reference the price history, don’t open suspicious emails, and don’t follow dubious links. If you follow these steps, your Black Friday will result not only in zero losses, but also in genuinely advantageous purchases.

What else to read on the topic of AI:

• Privacy settings in ChatGPT
• DeepSeek: configuring privacy and deploying a local version
• The hidden dangers of AI coding
• New types of attacks on AI-powered assistants and chatbots
• How phishers and scammers use AI

Kaspersky official blog – ​Read More

Some attacks smash the door open. LOLBins just borrow your keys and walk right in. 

They’re tricky because tools everyone trusts suddenly start doing things that don’t match their usual job; loading odd-looking modules, decoding files that shouldn’t need decoding, or quietly handing work off to hidden PowerShell scripts. At first glance it all feels normal, but a closer look shows a payload slowly being set up in the background. 

For analysts, the real challenge is noticing that shift before it grows into a full incident. 

Let’s take a closer look at what’s hiding behind LOLBin attacks, and how advanced SOC teams uncover them in minutes without much effort. 

What Are LOLBin Attacks? 

LOLBin attacks occur when threat actors repurpose legitimate Windows system binaries (rundll32, certutil, mshta, powershell, regsvr32, etc.) to carry out malicious actions. These tools are built into every system, signed by Microsoft, and widely used by normal applications, which is why attackers rely on them. 

Using LOLBins, adversaries can: 

• Load disguised or renamed DLLs 

• Decode or unpack payloads using built-in utilities 

• Trigger PowerShell or script execution indirectly 

• Execute code completely in memory 

• Blend malicious steps into routine system activity 

This approach lets attackers avoid dropping obvious malware and makes early-stage execution appear clean and legitimate. 

Why LOLBin Attacks Are a Real Risk for Businesses? 

The real problem isn’t the binaries themselves but how much visibility your SOC loses when attackers hide behind them. When malicious activity runs inside trusted system tools, the early signs of an intrusion become dramatically harder to catch. 

Here’s what makes them dangerous: 

• Normal on the surface: Activity is routed through tools the environment already trusts. 

• Minimal forensic evidence: In-memory execution leaves few files to investigate. 

• Weak signature coverage: Microsoft-signed binaries rarely trigger basic detection rules. 

• Extended dwell time: Attackers gain more space for lateral movement and credential access. 

• Harder investigations: Clean-looking events force analysts to dig deeper to find the real issue. 

• Higher SOC workload: The team must identify subtle behavior shifts instead of relying on clear indicators. 

This means attackers can establish footholds, unpack payloads, or run loaders while the environment still appears clean, leading to late detection and higher incident impact. 

The Fastest Way to Reveal LOLBin Abuse: How ANY.RUN Makes It Obvious 

LOLBin attacks only work when no one can see what’s really happening behind those trusted Windows binaries. ANY.RUN removes that advantage by showing analysts the full behavior in real time; not just the file name or the process label, but the actual actions taking place underneath. 

With ANY.RUN’s sandbox, “normal-looking” activity turns into something you can spot immediately: 

• Process behavior becomes clear at a glance: rundll32 loading a strange module, certutil decoding an unexpected file, mshta spawning hidden PowerShell… every unusual step is visible right away. 

• Parent–child chains tell the full story: Instead of digging through logs, you see exactly who launched what, and whether it fits normal usage patterns. 

• Command lines show the truth: Encoded strings, odd export calls, Temp-folder payloads, and hidden flags are exposed instantly. 

• In-memory actions are no longer invisible: Even when attackers avoid dropping files, the sandbox reveals decoded scripts, loader behavior, and execution flow. 

• Artifacts stay captured: Renamed DLLs, extracted archives, decrypted payloads, and cleanup attempts can all be reviewed without rushing or digging. 

• Analysis becomes interactive: Analysts can click deeper, replay events, and confirm suspicions in minutes instead of piecing everything together manually. 

Instead of guessing whether a trusted binary is being misused, ANY.RUN shows the exact behavior clearly, quickly, and with the context you need to act confidently. 

Real-Time LOLBin Attacks Revealed Inside ANY.RUN in Minutes 

Here are a few real LOLBin attacks captured and analyzed inside ANY.RUN. 
Take a look at how these techniques unfold in real time, and see how easily your team can expose the same behavior using interactive analysis. 

1. LOLBin RUNDLL32.EXE 

ATT&CK® Technique: T1218.011 – Rundll32 

What this attack is: 
A trusted Windows utility used to load and run a disguised module, letting attackers execute their payload under a legitimate process. 

See this RUNDLL32 attack exposed live inside ANY.RUN: 
→ Gh0st RAT delivered through rundll32 

Gh0st RAT launches the legitimate rundll32.exe, which then loads a disguised module named grgfrqe.rfg from an unusual directory. The file isn’t a typical DLL at first glance; the strange extension is intentionally chosen to bypass simple “.dll” rules and blend into the system. 

Once loaded, rundll32 calls an export named RAFlush and passes it a path to a temporary executable: C:UsersadminAppDataLocalTemphkjhn.exe. 

From there, the chain unfolds: 

• Load: rundll32 loads the renamed DLL (grgfrqe.rfg) 

• Invoke: The RAFlush export is executed 

• Drop/execute: The module drops, unpacks, or runs hkjhn.exe inside %Temp% 

• Cleanup: Temporary files are removed to reduce traces 

This is a typical LOLBin pattern: a trusted binary quietly executing hidden functionality while the malicious module stays disguised and difficult to catch without behavioral visibility. 

Use this ANY.RUN’s TI Lookup query to explore similar samples and collect IOCs: 

commandLine:”rundll32.exe*” 

2. LOLBin CERTUTIL.EXE 

ATT&CK® Technique: T1140 – Deobfuscate/Decode Files or Information 

What this attack is: 
A built-in Windows tool misused to decode, transform, or prepare hidden payloads before execution; all under the guise of a legitimate system operation. 

See this CERTUTIL attack exposed live inside ANY.RUN: 
→ PXAStealer decoding and unpacking files through certutil 

PXAStealer uses certutil.exe to quietly decode a disguised file named DA 성형외과 재무 보고서.pdf. Although it appears to be a harmless PDF, certutil converts it into Invoice.pdf, which is not a document at all but a RAR archive. 

The attack continues as a renamed instance of WinRAR, disguised as a JPEG image (부가가치세 영수증.jpg), unpacks the archive using the password 
iJbcsRBR84uUl9USIhj09PH0elalyHPJ. 

The execution flow looks like this: 

• Decode: certutil transforms the fake PDF into an archive 

• Extract: The disguised WinRAR instance unpacks it 

• Execute: The payload inside the archive is launched 

• Cleanup: Files are removed or hidden to minimize traces 

This combination, a trusted decoding tool + disguised content + hidden extraction, is a classic LOLBin chain designed to slip past basic detection and appear routine unless investigated behaviorally. 

Check out more sessions of this attack and gather related IOCs using this TI query: 

commandLine:”certutil.exe*-decode” 

3. LOLBin MSHTA.EXE 

ATT&CK® Technique: T1218.005 – Mshta 

What this attack is: 
A trusted Windows utility used to execute HTA-based scripts that trigger hidden PowerShell activity, enabling in-memory execution without leaving clear artifacts. 

See this MSHTA attack exposed live inside ANY.RUN: 
→ ReverseLoader executed through mshta + hidden PowerShell 

In this attack chain, mshta.exe launches an HTA file named gg.hta from the user’s desktop. The HTA isn’t a simple script; it contains obfuscated logic that immediately spawns a PowerShell process configured to stay out of sight. 

PowerShell is executed with: 

• -NoProfile 

• -WindowStyle Hidden 

• A Base64-encoded command decoded and passed into Invoke-Expression 

This allows the payload to run entirely in memory, without dropping a traditional file on disk. 

Here’s how the chain unfolds: 

• Deliver: The HTA file is delivered locally or through a link 

• Execute: mshta runs the HTA script as a trusted system tool 

• Decode & run: PowerShell decodes the Base64 string and executes the logic 

• Stealth: Hidden windows and in-memory execution conceal most traces 

This mshta + encoded PowerShell combination is a well-known method for quietly loading backdoors, RATs, and script-based loaders while appearing to use legitimate system components. 

Check out more sessions of similar attacks and gather relevant data using this TI query: 

commandLine:”mshta.exe*.hta” 

Strengthening Defenses Against LOLBin Techniques 

For SOC managers, stopping LOLBin abuse starts with improving how the team spots unusual behavior inside trusted system tools. These attacks don’t announce themselves, so the goal is to create clearer visibility and reduce the time analysts spend guessing what’s happening. 

Focus on behavior, not the binary: Even legitimate tools like rundll32, certutil, and mshta become suspicious when they load odd modules, decode files, or trigger hidden PowerShell. Building detections around these behaviors helps the team surface threats that signatures often miss. 

Give analysts a simple triage path: Most LOLBin alerts look harmless at first. A lightweight checklist, parent process, command line, execution path, and any decoding or script activity, keeps investigations consistent and prevents early-stage activity from slipping by. 

Use sandbox analysis to confirm suspicious cases quickly: Instead of piecing clues together from logs, ANY.RUN gives analysts the full picture in seconds: process chains, decoded content, dropped components, and in-memory activity. This cuts investigation time and helps the team act confidently. 

Add small policy controls where possible: Limiting execution from user-controlled folders or applying basic PowerShell restrictions reduces the surface attackers can exploit without disrupting normal operations. 

A few focused improvements like these help SOC managers turn LOLBin activity from a hidden risk into something the team can catch early and handle efficiently. 

About ANY.RUN 

ANY.RUN is a leading provider of interactive malware analysis and threat intelligence solutions, built to give SOC teams the visibility they need when traditional tools fall short.  

Today, 15,000+ organizations worldwide use ANY.RUN to speed up investigations, strengthen detection pipelines, and give their teams a clearer view of what’s really happening on their endpoints. 

SOC teams using ANY.RUN report measurable improvements, including: 

• 3× boost in SOC efficiency 

• 95% faster initial triage 

• Up to 58% more threats identified 

• 21-minute reduction in MTTR per incident 

Give your team the visibility they need: Try ANY.RUN now 

The post LOLBin Attacks Explained with Examples: Everything SOC Teams Need to Know  appeared first on ANY.RUN’s Cybersecurity Blog.

ANY.RUN’s Cybersecurity Blog – ​Read More

Today, Cisco Talos is introducing new capabilities for Snort3 users within Cisco Secure Firewall. These enhancements are designed to give you greater flexibility in how you manage, organize, and prioritize detection rules. They also make it easier to align SNORT® rules with your organization’s specific security needs.

The new “Severity” rule group

In Snort3, rule groups let you organize and manage detection rules according to specific criteria. Previously, only two top-level groups were available:

• Rule Category: groups rules by Snort2 categories such as FILE-OTHER, MALWARE-CNC, etc.
• MITRE ATT&CK: groups rules by attacker behaviors and techniques

These groups allow you to set a security level from 0 (all rules disabled) to 4 (all rules enabled).

The new Severity rule group introduces a third way to organize rules — by vulnerability severity, using CVSS scores. Rules are grouped as low, medium, high, or critical, allowing your team to prioritize detection based on the impact and urgency of vulnerabilities, rather than just category or behavior.

This makes it easier to focus attention and resources where they matter most.

Flexible rule group creation based on time range

With the Severity group, you can define how far back in time you want your coverage to extend:

Level	Coverage	Description
0	None	No rules enabled
1	Last 2 years	Focuses on recent, high-impact vulnerabilities
2	Last 5 years	Balanced coverage of recent and mid-term threats
3	Last 10 years	Broad coverage for long-lived environments
4	All	Includes all vulnerabilities detected to date

This approach gives you precise control over rule selection and volume. It helps optimize performance while ensuring your detection policies match your organization’s patching cycles, compliance requirements, and risk profile.

We’re also looking to develop more top-level groupings in the coming quarters. More details will be shared in due course.

What this means for your environment

Configuring Snort3 previously required enabling rules individually or applying a predefined ruleset and then tuning manually. We know this wasn’t the most time-efficient process, so the Snort analyst team worked to simplify it with the new features announced today.

You can now:

• Enable rule groups aligned with your own internal policies
• Scale configurations across multiple environments without managing individual rules
• Adjust detection depth easily by time range or severity level

These capabilities make it simpler to maintain consistent, targeted detection coverage — whether you’re running large, distributed networks or smaller environments with tailored security priorities.

Conclusion

The new Severity rule group and expanded rule group model give Snort3 users more flexibility and control.

By organizing rules based on vulnerability severity and timeframe, you can focus detection where it has the greatest impact, improving both efficiency and accuracy in threat management.

Cisco Talos Blog – ​Read More

Scaling as a managed security provider can be a mixed blessing. Growth comes with more revenue, but also with increasingly high demands related to maintaining SLAs, quality, and compliance. For MSSPs in healthcare, this pressure is intensified by regulations like HIPAA and NIS2, along with the striking cost of a single mistake. 

This was a challenge one of our clients, a mid-sized MSSP specializing in healthcare, had to face. As it expanded to support over a dozen hospitals, clinics, and labs with 2,000+ endpoints, their resources were thinning.  

We spoke with the organization’s SOC lead about how they were able to reshape their workflow with ANY.RUN’s solutions, and what brought them the most results. 

Core MSSP Challenges: Overload and Compliance 

The first topic we discussed was what the workflow was like initially and why the need for new solutions occurred. In their words, even with experts on board and acknowledged tools, occurring gaps were growing harder to fill: 

“It [the workflow] wasn’t that bad: we have a strong team and a SOAR platform by a well-known vendor. Teamwork was – and remains – our strong point. But as the client base grew, it became harder to maintain SLAs, which are pretty strict in healthcare. Tier 1 and 2 analysts were overwhelmed by an increased number of alerts coming from different customers.” 

The analysts had to deal with hundreds of emails and URLs reported by clients each week, and the verification process was mostly manual. Some multi-step phishing cases required up to 40 minutes of analysis, as they required multiple tools and resources, or even custom virtual machines. The need for better triage solutions and prioritization protocols intensified. 

Key challenges: 

• Slow MTTR across multiple customers 

• Tier 1 analysts indicated roughly 20% closure rates 

• Excessive escalations from Tier 1 to Tier 2 

• Lack of automation in triage 

• Struggles to maintain compliance 

Concerns that Come with Growth 

After discussion, the company leaders came up with a plan to enhance the processes: 

1. Introducing more automation in alert triage to reduce workload 

2. Obtain higher-quality threat data for faster decision-making 

3. Shifting from reactive to proactive defense 

Some team members expressed concerns about introducing a new solution: 

““What if we have to rebuild the workflow from scratch? What if automation fails to work as promised?” – these are some of the questions the analysts raised. So we had to be selective [when choosing a solution]. We needed something flexible and easy to integrate.” 

Immediate Improvements with ANY.RUN’s Interactive Sandbox 

The MSSP launched the streamlining process by adding just one solution to the stack at a time. The choice fell on ANY.RUN’s Interactive Sandbox, as it offered a unique approach to dynamic malware analysis: 

“It stood out among other options with interactivity. Automation is powerful, but not always enough. Interactivity offered more depth and understanding of malware.” 

The MSSP has been using the sandbox for one and a half years, mostly as a solution integrated in their SOAR. The automated mode helped effortlessly deal with overflowing low-priority incidents, even if they included multiple stages like hidden links, redirects, and CAPTCHAs. 

Just in the past few weeks, multiple phishing campaigns were prevented through interactive analysis. Two of them involved common office tools used in medical institutions, such as OneDrive and Jotform: 

In SOC lead’s words, concerns related to automation and integration didn’t turn out to be justified: 

“The integration worked much better than was expected. With minimal changes in the workflow, we achieved stronger results: Tier 1 analysts had far more capacity; analysis of both low- and high-priority incidents became easier. No manual VM unfolding, no tedious escalations.” 

More Results with Threat Intelligence  

As part of a scheduled assessment of their infrastructure, the team was also shopping for new sources of threat intelligence. After a two-week trial, they decided to fully implement ANY.RUN’s products into the existing workflows.  

“We were happy with the results ANY.RUN’s sandbox brought, so it made sense to support resources we’ve been using for a while with TI solutions from the same vendor.” 

Threat Intelligence Lookup and Threat Intelligence Feeds added a finishing touch to the new defense strategy. They brought outcomes like: 

• Real-time, high-confidence IOCs Better preventative measures 

• Broad threat monitoring Early detection of attacks 

• Threat context just a click away Fast enrichment of isolated artifacts 

• Behavioral data through sandbox analyses New detection rules 

• Automation via SOAR integration Effortless responses and ticket closure 

Together, these solutions enabled the SOC to proactively hunt and neutralize threats before they could impact client operations. 

Measurable Outcomes 

*Based on the company’s statistics after using ANY.RUN’s solutions 

Phishing Campaign Case and Successful Mitigation 

A recent incident illustrated the efficiency of the new workflow based on early detection and mitigation: 

“A couple of weeks ago, our analysts spotted a suspicious connection on a client endpoint. TI Lookup immediately showed that it’s tied to a known malicious C2 server.” 

“For further insights, they browsed other analyses and saw a threat sample featuring phishing. The sandbox then helped uncover the entire attack chain; and retrieved IOC were used to refine detection rules.” 

Meeting Healthcare’s High Compliance Bar  

Healthcare is a sector with real urgency and high regulatory demands. Acting as an MSSP in this industry requires auditability, transparency, and SLA adherence. The SOC lead noted that protocols and regulations that are common in healthcare industry became easier to fulfill with ANY.RUN: 

“Since we implemented new solutions, every investigation now comes with evidence and threat data, from MITRE tags to screenshots. This made reporting faster and extra work fell off our shoulders.” 

The MSSP is now driven by: 

• Faster triage across multiple customers 

• Proactive and scalable threat detection strategy 

• Decision-making supported by high-quality data 

• Audit-ready evidence aligned with industry regulations 

Conclusion: From Reactive to Proactive Defense  

By integrating ANY.RUN’s Interactive Sandbox, TI Lookup, and TI Feeds, this MSSP built a proactive defense system.  

“Needless to say, we still work hard every day, but ANY.RUN gave us the tools to manage our daily tasks more effectively. More clarity and quick access to reliable information made all the difference. It lightened our load without taking away in quality.” 

About ANY.RUN 

Built for modern MSSPs and Enterprises, ANY.RUN empowers analysts to deliver faster, deeper, and more transparent threat analysis for their clients. The Interactive Sandbox exposes full attack behavior, from process execution to network activity, enabling analysts to investigate incidents in real time and make confident, data-driven decisions. 

Cloud-based and ready out of the box, ANY.RUN supports Windows, Linux, and Android environments, streamlining multi-tenant operations without complex setup. Integrated Threat Intelligence Lookup and TI Feeds provide continuously updated, automation-ready IOCs for better detection, response, and reporting across all client environments. 

See how ANY.RUN can elevate your MSSP: start a 14-day trial today 

The post Healthcare MSSP Cuts Phishing Triage by 76% and Launches Proactive Defense with ANY.RUN  appeared first on ANY.RUN’s Cybersecurity Blog.

ANY.RUN’s Cybersecurity Blog – ​Read More

Cybersecurity researchers have revealed a new attack method targeting AI browsers, which they refer to as AI sidebar spoofing. This attack exploits users’ growing habit of blindly trusting instructions they get from artificial intelligence. The researchers successfully implemented AI sidebar spoofing against two popular AI browsers: Comet by Perplexity and Atlas by OpenAI.

Initially, the researchers used Comet for their experiments, but later confirmed that the attack was viable in the Atlas browser as well. This post uses Comet as an example when explaining the mechanics of AI sidebar spoofing, but we urge the reader to remember that everything stated below also applies to Atlas.

How do AI browsers work?

To begin, let’s wrap our heads around AI browsers. The idea of artificial intelligence replacing, or at least transforming the familiar process of searching the internet began to generate buzz between 2023 and 2024. The same period saw the first-ever attempts to integrate AI into online searches.

Initially, these were supplementary features within conventional browsers — such as Microsoft Edge Copilot and Brave Leo — implemented as AI sidebars. They added built-in assistants to the browser interface for summarizing pages, answering questions, and navigating sites. By 2025, the evolution of this concept ushered in Comet from Perplexity AI — the first browser designed for user-AI interaction from the ground up.

This made artificial intelligence the centerpiece of Comet’s user interface, rather than just an add-on. It unified search, analysis, and work automation into a seamless experience. Shortly thereafter, in October 2025, OpenAI introduced its own AI browser, Atlas, which was built around the same concept.

Comet’s primary interface element is the input bar in the center of the screen, through which the user interacts with the AI. It’s the same with Atlas.

Besides, AI browsers allow users to engage with the artificial intelligence right on the web page. They do this through a built-in sidebar that analyzes content and handles queries — all without having the user leave the page. The user can ask the AI to summarize an article, explain a term, compare data, or generate a command while remaining on the current page.

This level of integration conditions users to take the answers and instructions provided by the built-in AI for granted. When an assistant is seamlessly built into the user interface and feels like a natural part of the system, most people rarely stop to double-check the actions it suggests.

This trust is precisely what the attack demonstrated by the researchers exploits. A fake AI sidebar can issue false instructions — directing the user to execute malicious commands or visit phishing websites.

How did the researchers manage to execute the AI sidebar spoofing attack?

The attack starts with the user installing a malicious extension. To do its evil deeds, it needs permissions to view and modify data on all visited sites, as well as access to the client-side data storage API.

All of these are quite standard permissions; without the first one — no browser extension will work at all. Therefore, the chances that the user will get suspicious when a new extension requests these permissions are almost zero. You can read more about browser extensions and the permissions they request in our post Browser extensions: more dangerous than you think.

Once installed, the extension injects JavaScript into the web page and creates a counterfeit sidebar that looks strikingly similar to the real thing. This shouldn’t raise any red flags with the user: when the extension receives a query, it talks to the legitimate LLM and faithfully displays its response. The researchers used Google Gemini in their experiments, though OpenAI’s ChatGPT likely would have worked just as well.

The fake sidebar can selectively manipulate responses to specific topics or key queries set in advance by the potential attacker. This means that in most cases, the extension will simply display legitimate AI responses, but in certain situations it will display malicious instructions, links, or commands instead.

How realistic is the scenario where an unsuspecting user installs a malicious extension capable of the actions described above? Experience shows it is highly probable. On our blog, we’ve repeatedly reported on dozens of malicious and suspicious extensions that successfully make it into the official Chrome Web Store. This continues to occur despite all the security checks conducted by the store and the vast resources at Google’s disposal. Read more about how malicious extensions end up in official stores in our post 57 shady Chrome extensions clock up six million installs.

Consequences of AI sidebar spoofing

Now let’s discuss what attackers can use a fake sidebar for. As noted by the researchers, the AI sidebar spoofing attack offers potential malicious actors ample opportunities to cause harm. To demonstrate this, the researchers described three possible attack scenarios and their consequences: crypto-wallet phishing, Google account theft, and device takeover. Let’s examine each of them in detail.

Using a fake AI sidebar to steal Binance credentials

In the first scenario, the user asks the AI in the sidebar how to sell their cryptocurrency on the Binance crypto exchange. The AI assistant provides a detailed answer that includes a link to the crypto exchange. But this link doesn’t lead to the real Binance site — it takes you to a remarkably convincing fake. The link points to the attacker’s phishing site, which uses the fake domain name binacee.

Next, the unsuspecting user enters their Binance credentials and the code for two-factor authentication, if needed. After this, the attackers gain full access to the victim’s account and can siphon off all funds from their crypto wallets.

Using a fake AI sidebar to take over a Google account

The next attack variation also begins with a phishing link — in this case, to a fake file-sharing service. If the user clicks the link, they’re taken to a website where the landing page prompts them to sign in with their Google account right away.

After the user clicks this option, they’re redirected to the legitimate Google login page to enter their credentials there, but then the fake platform requests full access to the user’s Google Drive and Gmail.

If the user fails to scrutinize the page, and automatically clicks Allow, they grant attackers permissions for highly dangerous actions:

• Viewing their emails and settings.
• Reading, creating, and sending emails from their Gmail account.
• Viewing and downloading all the files they store in Google Drive.

This level of access gives the cybercriminals the ability to steal the victim’s files, use services and accounts linked to that email address, and impersonate the account owner to disseminate phishing messages.

Reverse shell initiated through a fake AI-generated utility installation guide

Finally, in the last scenario, the user asks the AI how to install a certain application; the Homebrew utility was used in the example, but it could be anything. The sidebar shows the user a perfectly reasonable, AI-generated guide. All steps in it look plausible and correct up until the final stage, where the utility installation command is replaced with a reverse shell.

If the user follows the AI’s instructions by copying and pasting the malicious code into the terminal and then running it, their system will be compromised. The attackers will be able to download data from the device, monitor activity, or install malware and continue the attack. This scenario clearly demonstrates that a single replaced line of code in a trusted AI interface is capable of fully compromising a device.

How to avoid becoming a victim of fake AI-sidebars

The AI sidebar spoofing attack scheme is currently only theoretical. However, in recent years attackers have been very quick to turn hypothetical threats into practical attacks. Thus, it’s quite possible that some malware creator is already hard at work on a malicious extension using a fake AI-sidebar, or uploading one to an official extension store.

Therefore, it’s important to remember that even a familiar browser interface can be compromised. And even if instructions look convincing and come from the in-browser AI assistant, you shouldn’t blindly trust them. Here’s some final tips to help you avoid falling victim to an attack involving fake AI:

• When using AI assistants, carefully check all commands and links before following the AI’s recommendations.
• If the AI recommends executing any programming code, copy it and find out what it does by pasting it into a search engine in a different, non-AI browser.
• Don’t install browser extensions — AI or otherwise — unless absolutely necessary. Regularly clean up and delete any extensions you no longer use.
• Before installing an extension, read the user reviews. Most malicious extensions rack up heaps of scathing reviews from duped users long before store moderators get around to removing them.
• Before entering credentials or other confidential information, always check that the website address doesn’t look suspicious or contain typos. Pay attention to the top-level domain, too: it should be the official one.
• Use Kaspersky Password Manager to store passwords. If it doesn’t recognize the site and doesn’t automatically offer to fill in the login and password fields, this is a strong reason to ask yourself if you might be on a phishing page.
• Install a reliable security solution that will alert you to suspicious activity on your device and prevent you from visiting a phishing site.

What other threats await you in browsers ­— AI-powered or regular:

• The pros and cons of AI-powered browsers
• Taking the biscuit: why hackers like cookies so much
• Turning purple: how visited links threaten your privacy
• Privacy under attack: nasty surprises in Chrome, Edge, and Firefox
• Dangerous browser extensions

Kaspersky official blog – ​Read More