Some scammers are selling ETA documents at exaggerated prices, and others are after your personal and financial data.

Since January 8, 2025, travelers from most countries, including the US, Australia, and Canada have to apply for an Electronic Travel Authorisation (ETA) for visa free travel to the UK.

You can apply for an Electronic Travel Authorisation using the ETA App, or via an online form.

When you apply for a UK ETA you have to pay an application fee of £10 ($13.50), provide your contact and passport details, a valid passport photo, and answer a set of questions about suitability.

But as often happens when new regulations take effect, scammers are quick to exploit unsuspecting travelers by charging inflated fees or collecting personal information.

There are some that will get you an actual ETA, but at exorbitant prices, and some that you’ll never hear from again after you have made a payment.

Some scammers will promise to charge extra for guiding you quickly through the process, such as in this example we saw.

But there is absolutely no need to do that. Most ETA applicants receive an automatic decision within minutes when applying through the official UK ETA app, according to the UK government. However, some applications may take up to three working days to be processed, so, it’s a good idea to apply at least a few days before travel.

If you have applied for an ETA through an unofficial channel, or you have doubts about the validity of your ETA, you can check using the “Check eTA status” tool on the official UK government website. You will need your eTA reference number (found in your confirmation email) and the details of the passport used for the application.

The UK is not the only country targeted in this type of scam. One phishing scam targeted Canadian ETA applicants using fake websites closely mimicking the official Government of Canada eTA application site. Victims were lured to apply for travel authorization, providing personal data such as full name, passport number, and more. They were then asked to pay CAD $100, far above the official CAD $7 fee, to obtain the authorization.

Some AI search aids—I’m looking at you Gemini—will tell you there is an online “Check eTA status tool” which is only true for Canada. There is no such thing for the UK.

And searching for that tool brought me to another site that charges overpriced rates for an ETA.

Regarding the UK, once your application is approved, your ETA is electronically linked to the passport you used in your application. You will receive an email confirmation and a 16-digit ETA reference number from UK Visas and Immigration (UKVI), usually within three working days of applying. There is no need to present any document besides your passport when entering the UK.

**How to avoid ETA scams**

The risks of applying for an ETA through other sites may vary and range from paying more than necessary to identity theft. Here are several tips to stay safe when applying for your ETA.

*   Stick to the official .gov.uk site, or use the official ETA app to make your application.
*   Be wary of websites charging significantly higher fees than official amounts.
*   Avoid clicking on sponsored advertisements or top search results without verifying their legitimacy.  
    
*   Do not share personal or payment data on suspicious or unknown sites.
*   When contacted by email or phone about ETAs or visas, verify the contact through official government sources.
*   Use an up-to-date real-time security solution that includes web protection.

**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.

 Travelers to the UK targeted in ETA scams 

A list of topics we covered in the week of August 25 to August 31 of 2025

August 28, 2025 - The FCC has disconnected over a thousand voice operators from the public telephone network for not doing their part to stop robocallers.

August 28, 2025 - Anthropic—maker of AI coding chatbot Claude—says cybercriminals have abused Claude to automate and orchestrate sophisticated attacks.

August 27, 2025 - To reduce the number of harmful apps targeting Android users, Google is making some changes.

August 27, 2025 - TheTruthSpy is at it again. A security researcher has discovered a flaw in the Android-based stalkerware that allows anyone to compromise any record in the system.

August 27, 2025 - Researchers have found 77 malicious apps in the official Google Play Store, ranging from adware to state of the art banking Trojans.

 A week in security (August 25 &#8211; August 31) 

Microsoft is rolling out a feature that defaults to saving your documents to the cloud. Consumers are divided.

Microsoft has revealed it plans to automatically save all Word document to the cloud. The feature is currently only available to Microsoft 365 Insiders, although it’s likely to expand this to all users in the future.

Microsoft proudly announced:

> “We are modernizing the way files are created and stored in Word for Windows! Now you don’t have to worry about saving your documents: Anything new you create will be saved automatically to OneDrive or your preferred cloud destination.”

The options this feature provides already existed, but Microsoft is changing the default save location for Word documents on Windows.

And Word is just the start:

> “Similar functionality is coming to Excel for Windows and PowerPoint for Windows later this year.”

For those of us that have lost hours of work because we forgot to press Save while we were still writing, this could be a blessing at times. But we can already enable AutoSave and choose OneDrive as an option. So what’s new?

The key change with the new plan is that Word now saves all new documents automatically to the cloud by default, before you even assign a file name. This removes the step where users had to manually initiate the process—the first save, so to speak. Whether you saved the first file locally or in the cloud, until now AutoSave used to activate only after your manual save to OneDrive. Now, Word creates every new document directly in the cloud and turns on AutoSave immediately from the start.

And the advantages are clear. No more forgetting to save, and you can share the document with someone else for collaboration purposes. It also means you can work on the document from any computer, as long as you can login with your Microsoft credentials.

But that last advantage could also be a pitfall. Anyone with your credentials can access all the documents you saved to OneDrive. And even though I realize this may sound alarmist, I have to point out that breaches happen, credentials get stolen, and more documents will be found when this is the default setting.

Users are afraid that with AI integration (Copilot) and saving documents to the cloud, their work will be used to train artificial intelligence (AI) or that this is a Microsoft scheme to sell more cloud storage.

But most of the disgruntled users are saying that big tech’s habit to turn things on by default is annoying, to put it mildly.

**Some pointers**

So, what if you don’t want all your documents saved in the cloud? Or you want to change any of the other default settings that come with this feature?

Other things to remember:

*   You still can’t close a document before saving if you ever need it again. If you close the document before saving, a dialog will appear asking you whether you want to **Discard** or **Keep** it. When you close an empty document, the system discards it without asking for confirmation.
*   If you start a new Word session while another session is running, Word does not automatically save the new file due to a known issue.
*   If you disable the **Show the Start screen when this application starts** setting, Word won’t automatically save the first file you create after launching it.

**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.

 Microsoft wants to automatically save your Word docs to the cloud 

The FCC has disconnected over a thousand voice operators from the public telephone network for not doing their part to stop robocallers.

Everyone hates robocalls. However, it’s difficult to track down all the scammers and spammers that make them, so the Federal Communications Commission (FCC) has taken another approach: it just disconnected over a thousand voice operators from the public telephone network for not doing their part to stop the scourge.

This week, the Commission’s Enforcement Bureau removed over 1,200 voice service providers from its Robocall Mitigation Database (RMD). Created in 2020, this database is a ledger with records proving that telephony operators are taking measures to stop robocalls routing their calls through their networks. Removal from the database prevents other operators from taking a service provider’s traffic, effectively cutting it off from the US phone network.

**Shaken and stirred**

There’s a long road leading to this point, starting with the development of the STIR/SHAKEN protocol. Secure Telephone Identity Revisited (STIR) is a standard for ensuring that the caller ID showing up on your phone is legit. Signature-based Handling of Asserted information using toKENs (SHAKEN) is the technical tooling that lets telephony providers use the standard on their network.

Large voice telephony providers had to implement STIR/SHAKEN by June 30 2021 under the TRACED Act of 2019. Smaller providers got an extension. The RMD tracks which providers are using this system.

The Commission has been tightening the screws on companies that didn’t comply with the Act. In December, it announced that it might remove up to 2,411 companies if they couldn’t give a good reason for why they didn’t have up-to-date filings in the database.

On August 6, it began delivering on its promise, removing 185 voice providers from the database after they were found to be an originator or gateway provider for robocalls, or after they didn’t co-operate with the traceback procedures used to trace those calls.

The FCC also has support at a state level. In early August, 51 attorneys general launched Operation Robocall Roundup, which sent letters to 37 voice operators putting them on notice about illegal robocalls using their networks.

According to Commission Chair Brendan Carr:

> “Robocalls are an all-too-common frustration — and threat — to Americans \[sic\] households. The FCC is doing everything in its power to fight back against these malicious and illegal calls. Providers that fail to do their duty when it comes to stopping these calls have no place in our networks. We’re taking action and we will continue to do so.”

**Protection isn’t guaranteed**

This is great, as far as it goes, but STIR/SHAKEN only works on IP-based phone networks (those that use the same protocol that the internet uses to move their digitized voice data around). Legacy phone networks that don’t use IP, such as in some rural areas, can’t use the technology. Those will fade over time, though.

Perhaps more importantly, the STIR/SHAKEN rules apply to US providers only, and it’s cheap for overseas providers to reach you. So overseas robocallers can still get to you easily via non-US operators while spoofing caller IDs.

Finally, STIR/SHAKEN only proves that the number showing up on your phone is the number that’s actually calling. The person using that number could still be a scammer.

So, you still need to do a little legwork of your own to minimize robocalls. Network providers (typically the larger ones) often run their own network analytics to root out robocallers. They frequently bundle such services. You can also set your phone to send all calls from numbers not in your contact list straight to voicemail.

If you use a landline, you can connect call blocking devices to your phone. Those use a variety of tricks, including messages that require a caller to press a specific number before the call goes through. The more action you take at your end, the more likely you are to block out automated nuisances.

**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.

 &#8220;No place in our networks&#8221;: FCC hangs up on thousands of voice operators in robocall war 

Anthropic—maker of AI coding chatbot Claude—says cybercriminals have abused Claude to automate and orchestrate sophisticated attacks.

Anthropic—the company behind the widely renowned coding chatbot, Claude—says it uncovered a large-scale extortion operation in which cybercriminals abused Claude to automate and orchestrate sophisticated attacks.

The company issued a Threat Intelligence report in which it describes several instances of Claude abuse. In the report it states that:

> “Cyber threat actors leverage AI—using coding agents to actively execute operations on victim networks, known as vibe hacking.”

This means that cybercriminals found ways to exploit vibe coding by using AI to design and launch attacks. Vibe coding is a way of creating software using AI, where someone simply describes what they want an app or program to do in plain language, and the AI writes the actual code to make it happen.

The process is much less technical than traditional programming, making it easy and fast to build applications, even for those who aren’t expert coders. For cybercriminals this lowers the bar for the technical knowledge needed to launch attacks, and helps the criminals to do it faster and at a larger scale.

Anthropic provides several examples of Claude’s abuse by cybercriminals. One of them was a large-scale operation which potentially affected at least 17 distinct organizations in just the last month across government, healthcare, emergency services, and religious institutions.

The people behind these attacks integrated the use of open source intelligence tools with an “unprecedented integration of artificial intelligence throughout their attack lifecycle.”

This systematic approach resulted in the compromise of personal records, including healthcare data, financial information, government credentials, and other sensitive information.

The primary goal of the cybercriminals is the extortion of the compromised organizations. The attacker created ransom notes to compromised systems demanding payments ranging from $75,000 to $500,000 in Bitcoin. But if the targets refuse to pay, the stolen personal records are bound to be published or sold to other cybercriminals.

Other campaigns stopped by Anthropic involved North Korean IT worker schemes, Ransomware-as-a-Service operations, credit card fraud, information stealer log analysis, a romance scam bot, and a Russian-speaking developer using Claude to create malware with advanced evasion capabilities.

But the case in which Anthropic found cybercriminals attack at least 17 organizations represents an entirely new phenomenon where the attacker used AI throughout the entire operation. From gaining access to the target’s systems to writing the ransomware notes—for every step Claude was used to automate this cybercrime spree.

Anthropic deploys a Threat Intelligence team to investigate real world abuse of their AI agents and works with other teams to find and improve defenses against this type of abuse. They also share key findings of the indicators with partners to help prevent similar abuse across the ecosystem.

Anthropic did not name any of the 17 organizations, but it stands to reason we’ll learn who they are sooner or later. One by one, when they report data breaches, or as a whole if the cybercriminals decide to publish a list.

Data breaches of organizations that we’ve given our data to happen all the time, and that stolen information is often published online. Malwarebytes has a free tool for you to check how much of your personal data has been exposed—just submit your email address (it’s best to give the one you most frequently use) to our free Digital Footprint scanner and we’ll give you a report and recommendations.

 Claude AI chatbot abused to launch “cybercrime spree” 

To reduce the number of harmful apps targeting Android users, Google is making some changes.

To reduce the number of harmful apps targeting Android users, Google has announced that certified Android devices will require all apps to be registered by verified developers in order to be installed.

But this new measure is not just about malware that’s found on the Google Play Store, it’s mainly about sideloaded apps (apps downloaded from outside the official Google Play Store).

Since August 31, 2023, apps on the Play Store already were subject to a D-U-N-S (Data Universal Numbering System) number requirement. Google says this has helped reduce the number of cybercriminals exploiting anonymity to distribute malware, commit financial fraud, and steal sensitive data.

To broaden this success, Google intends to start sending out invitations gradually starting October 2025, before opening it up to all developers in March 2026. In September 2026, the requirements go into effect in Brazil, Indonesia, Singapore, and Thailand. At this point, any app installed on a certified Android device in these regions must be registered by a verified developer. The requirements will then be rolled out globally.

This initiative, branded as ‘Developer verification,’ aims to combat the widespread problem of malware from sideloaded apps. Google says its research shows that 50 times more malware comes from sideloaded sources than from Google Play itself.

So, the new rules extend to everyone distributing Android apps, including those hosting them on third-party app stores or offering APK downloads directly. For developers who distribute their apps solely through the Google Play Store there will not be much of a change.

Yet, while legitimate developers will tell you how hard it is to get their apps accepted into the Google Play Store, cybercriminals manage to sneak in their malicious apps anyway.

For a full understanding of the new requirement, we’ll need to explain what “certified Android devices” are.

A definition for a certified Android device is: an Android product—such as a smartphone, tablet, smart TV, or streaming box—that has passed a rigorous series of Google security, compatibility, and performance tests, and is officially approved by Google. Certified devices run an official version of Android and have access to Google apps and the Play Store. Uncertified devices often lack these and may not receive updates or proper security support.

This is important to know because not all Android malware is limited to phones. Take for example, the BadBox botnet which also affects devices like TV streaming boxes, tablets, and smart TVs.

In practice, a certified device encompasses all mainstream devices from Samsung, Xiaomi, Motorola, OnePlus, Oppo, Vivo, and the Google Pixel line.

Reportedly, non-certified devices are those from Huawei, Amazon Fire tablets, and a set of Chinese TV boxes and smartphones that use heavily modified OS images.

Google encourages all developers to sign up for early access as the best way to prepare and stay informed.

>  “Early participants will also get:
> 
> *   An invitation to an exclusive community discussion forum.
> *   Priority support for these new requirements.
> *   The chance to provide feedback and help us shape the experience.”

Whether these controls will be effective largely depends on enforcement and public awareness, but Google feels it marks real progress toward a safer mobile ecosystem. Let us know how you feel about this in the comments.

**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.

 Developer verification: a promised lift for Android security 

TheTruthSpy is at it again. A security researcher has discovered a flaw in the Android-based stalkerware that allows anyone to compromise any record in the system.

TheTruthSpy is at it again. A security researcher has discovered a flaw in the Android-based stalkerware that allows anyone to compromise any record in the system.

TheTruthSpy stalkerware is designed to be installed surreptitiously on a victim’s Android phone. It then monitors that phone’s activities and sends the information it gathers back to a central server. On Monday, TechCrunch revealed—not for the first time—that the servers are vulnerable to attack. It found that anyone can reset the password of any account on the app, meaning they could hijack anyone’s data.

The security researcher, Swarang Wade, demonstrated the vulnerability to TechCrunch by changing the passwords on several tests accounts. The publication isn’t revealing exactly how it was done, to prevent anyone from abusing the flaw.

TheTruthSpy gathers a lot of data about its victims. It provides the person who installed it with information about what calls or texts were made or received on the victim’s phone, and its location (harvested from the GPS), along with activities associated with messaging apps and files.

This isn’t the first time that TheTruthSpy has suffered from security issues:

*   In 2018, a hacker gained access to all its servers.
*   In 2019, it exposed pictures of kids taken with infected phones online.
*   In 2024, stalkerware researchers found an unpatched flaw that rendered all data on the app’s servers accessible.

This would all be very bad if people using the app knew that they were doing so, and that their personal usage data was stored online. But many of them are oblivious to the fact.

TheTruthSpy’s vendor, Vietnam-based 1Byte Software, warns that people must obtain consent before installing the app on someone else’s phone. However it also specifically advertises ‘stealth mode’, which makes it “completely invisible to users on phones/tablets where it’s installed.”

The software’s website touts its ability to spy on phone users as a way for parents to monitor and protect their children. That raises its own ethical questions, especially given the multiple data leaks. But that isn’t its only use. Abusers will use apps like these to monitor their current or ex-partners, or other stalking targets.

Once a victim has this installed on their phone without their knowledge, the installer can monitor their photos, social media interactions, emails, and internet browsing history. It will also record audio and log keystrokes without them being aware.

Van (Vardy) Thieu, owner of 1Byte Software, told TechCrunch that its source code was lost. He claimed to be building a new version from scratch, although TechCrunch’s reporters found that it was using the same vulnerable software library as the older version.

The software’s multiple bugs demonstrate just how dangerous it is to put this – or indeed any stalkerware app – on someone’s phone. The operators of these apps are often difficult to track down and hold accountable for their security issues.

**How to check if you have stalkerware on your phone**

What can you do if you suspect your phone might be infected with stalkerware? We think TechCrunch’s guide deserves a mention here, as does The Coalition Against Stalkerware, of which Malwarebytes is a founding member. The latter includes per-country links to organizations that help victims of domestic violence.

It is good to keep in mind however that by removing any stalkerware-type app, you will alert the person spying on you that you know the app is there.

Because the apps install under a different name and hide themselves from the user, it can be hard to find and remove them. That is where Malwarebytes for Android can help you.

1.  Open Malwarebytes on your Android.
2.  Open the app’s dashboard
3.  Tap **Scan** **now**
4.  It may take a few minutes to scan your device.

 If malware is detected you can act on it in the following ways:

*   **Uninstall**. The threat will be deleted from your device.
*   **Ignore Always**. The file detection will be added to the Allow List, and excluded from future scans. Legitimate files are sometimes detected as malware. We recommend reviewing scan results and adding files to Ignore Always that you know are safe and want to keep.
*   **Ignore Once**: A file has been detected as a threat, but you are not sure whether to add it to your Allow List or delete. This option will ignore the detection this time only. It will be detected as malware on your next scan.

**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.

 More vulnerable stalkerware victims&#8217; data exposed in new TheTruthSpy flaw 

Researchers have found 77 malicious apps in the official Google Play Store, ranging from adware to state of the art banking Trojans.

Google has removed 77 malicious apps from the Google Play Store. Before they were removed, researchers at ThreatLabz discovered the apps had been installed over 19 million times.

One of the malware families discovered by the researchers is a banking Trojan known as Anatsa or TeaBot. This banking Trojan is a highly sophisticated Android malware, which focuses on stealing banking and cryptocurrency credentials.

Anatsa is a classic case of mobile malware rapidly adapting to security research progress. Its stealth tactics, exploitation of accessibility permissions, and ability to shift between hundreds of financial targets make it an ongoing threat for Android users worldwide.

Also found by the researchers were several types of adware. However, the largest chunk of malicious apps belonged to the Joker malware family, which is notorious for its stealthy behavior. It steals SMS messages, contacts, device info, and enrolls victims in unwanted premium services, which can result in financial losses.

The malware is installed like this:

*   It gets added to the Play Store as a benign app with useful and sought-after functionality (e.g. document readers, health trackers, keyboards, and photo apps).
*   Once installed, the app acts as a “dropper” which connects to a remote server for instructions and additional payloads, which often ends in the installation of information stealers.
*   Anatsa—specifically—uses several methods to avoid detection, such as a well-known Android APK ZIP obfuscator, and downloading each new chunk of code with a separate DES key.

Google says it picked up on the flaws and protected against these malware infections before the researchers published their report.

As a consequence, Google Play Protect may send users of the removed apps a push notification, giving them the option to remove the app from their device.

But don’t let that be your only line of defense. We found that Android users are more careful than iPhone users. Let’s keep that up!

**How to protect your Android from malicious apps**

Just because something is in the Google Play Store, there is no guarantee that it will remain a non-malicious app. So here are a few extra measures you can take:

*   Always check what permissions an app is requesting, and don’t just trust an app because it’s in the official Play Store. Ask questions such as: Do the permissions make sense for what the app is supposed to do? Why did necessary permissions change after an update? Do these changes make sense?
*   Occasionally go over your installed apps and remove any you no longer need.
*   Make sure you have the latest available updates for your device, and all your important apps (banking, security, etc.)
*   Protect your Android with security software. Your phone needs it just as much as your computer.

Malwarebytes for Android detects Anatsa as Trojan.Banker.CPL.

**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.

 77 malicious apps removed from Google Play Store 

Prompt injection attacks could be coming to an AI browser near you. Read on to understand what these attacks do and how to stay safe.

Artificial Intelligence (AI) browsers are gaining traction, which means we may need to start worrying about the potential dangers of something called “prompt injection.”

Large language models (LLMs)—like the ones that power AI chatbots including ChatGPT, Claude, and Gemini—are designed to follow “prompts,” which are the instructions and questions that people provide when looking up info or getting help with a topic. In a chatbot, the questions you ask the AI are the “prompts.” But AI models aren’t great at telling apart the types of commands that are meant for their eyes only (for example, hidden background rules that come directly from developers, like “don’t write ransomware“) from the types of requests that come from users.

To showcase the risks here, the web browser developer Brave—which has its own AI assistant called Leo—recently tested whether it could trick an AI browser into reading dangerous prompts that harm users. And what the company found caused alarm, as they wrote in a blog this week:

> “As users grow comfortable with AI browsers and begin trusting them with sensitive data in logged in sessions—such as banking, healthcare, and other critical websites—the risks multiply. What if the model hallucinates and performs actions you didn’t request? Or worse, what if a benign-looking website or a comment left on a social media site could steal your login credentials or other sensitive data by adding invisible instructions for the AI assistant?”

Prompt injection, then, is basically a trick where someone inserts carefully crafted input in the form of an ordinary conversation or data, to nudge or outright force an AI into doing something it wasn’t meant to do.

What sets prompt injection apart from old-school hacking is that the weapon here is language, not code. Attackers don’t need to break into servers or look for traditional software bugs, they just need to be clever with words.

For an AI browser, part of the input is the content of the sites it visits. So, it’s possible to hide indirect prompt injections inside web pages by embedding malicious instructions in content that appears harmless or invisible to human users but is processed by AI browsers as part of their command context.

Now we need to define the difference between an AI browser and an agentic browser. An AI browser is any browser that uses artificial intelligence to assist users. This might mean answering questions, summarizing articles, making recommendations, or helping with searches. These tools support the user but usually need some manual guidance and still rely on the user to approve or complete tasks.

But, more recently, we are seeing the rise of agentic browsers, which are a new type of web browser powered by artificial intelligence, designed to do much more than just display websites. These browsers are designed to actually take over entire workflows, executing complex multi-step tasks with little or no user intervention, meaning they can actually use and interact with sites to carry out tasks for the user, almost like having an online assistant. Instead of waiting for clicks and manual instructions, agentic browsers can navigate web pages, fill out forms, make purchases, or book appointments on their own, based on what the user wants to accomplish.

For example, when you tell your agentic browser, “Find the cheapest flight to Paris next month and book it,” the browser will do all the research, compare prices, fill out passenger details, and complete the booking without any extra steps or manual effort—provided it has all the necessary details of course, which are part of the prompts the user feeds the agentic browser.

Are you seeing the potential dangers of prompt injections here?

What if my agentic browser gets new details while visiting a website? I can imagine criminals setting up a website with extremely competitive pricing just to attract visitors, but the real goal is to extract the payment information which the agentic browser needs to make purchases on your behalf. You could end up paying for someone else’s vacation to France.

During their research, Brave found that Perplexity’s Comet has some vulnerabilities which “underline the security challenges faced by agentic AI implementations in browsers.”

The vulnerabilities allow an attack based on indirect prompt injection, which means the malicious instructions are embedded in external content (like a website, or a PDF) that the browser AI assistant processes as part of fulfilling the user’s request. There are various ways of hiding that malicious content from a casual inspection. Brave uses the example of white text on a white background which AI browsers have no problem reading and a human would not see without closer inspection.

To quote a user on X:

> “You can literally get prompt injected and your bank account drained by doomscrolling on reddit”

To prevent this type of prompt injection, it is imperative that agentic browsers understand the difference between user-provided instructions and web content processed to fulfill the instructions and treat them accordingly.

Perplexity has attempted twice to fix the vulnerability reported by Brave, but it still hasn’t fully mitigated this kind of attack as of the time of this reporting.

**Safe use of agentic browsers**

While it’s always tempting to use the latest gadgets this comes with a certain amount of risk. To limit those risks when using agentic browsers you should:

*   **Be cautious with permissions:** Only grant access to sensitive information or system controls when absolutely necessary. Review what data or accounts the agentic browser can access and limit permissions where possible.
*   **Verify sources before trusting links or commands:** Avoid letting the browser automatically interact with unfamiliar websites or content. Check URLs carefully and be wary of sudden redirects or unexpected input requests.
*   **Keep software updated:** Ensure the agentic browser and related AI tools are always running the latest versions to benefit from security patches and improvements against prompt injection exploits.
*   **Use strong authentication and monitoring:** Protect accounts connected to agentic browsers with multi-factor authentication and review activity logs regularly to spot unusual behavior early.
*   **Educate yourself about prompt injection risks:** Stay informed on the latest threats and best practices for safe AI interactions. Being aware is the first step to preventing exploitation.
*   **Limit sensitive operations automation:** Avoid fully automating high-stakes transactions or actions without manual review. Agentic browsers should assist, but critical decisions benefit from human oversight. For example: limit the amount of money it can spend without your explicit permission or always let it ask you to authorize payments.
*   **Report suspicious behavior:** If an agentic browser acts unpredictably or asks for strange permissions, report it to the developers or security teams immediately for investigation.

**We don’t just report on threats—we remove them**

Cybersecurity risks should never spread beyond a headline. Keep threats off your devices by downloading Malwarebytes today.

 AI browsers could leave users penniless: A prompt injection warning 

A list of topics we covered in the week of August 18 to August 24 of 2025

August 22, 2025 - Grok AI chats that users wanted to share with individual people were in fact shared with the broader web and searchable by everyone.

August 21, 2025 - Apple has released security updates to patch a zero-day vulnerability tracked as CVE-2025-43300 for all platforms

August 21, 2025 - Google has settled a lawsuit against YouTube for $30 million but did not admit collecting the data of minors for targeted advertising.

August 19, 2025 - Startups are ready to bring AI powered toys to the market as an alternative for screen time. But is that really progress?

August 18, 2025 - Meta has introduced Instagram Map. How can you control what others can see about your location? An explainer.

Source

Malwarebytes