A report by Incogni has revealed that two-thirds of popular AI Chrome extensions could wreak havoc on user security if turned malicious. The report draws to the importance of understanding the trade-offs between convenience vs user privacy and security when using these extensions.
“AI Chrome extensions offer undeniable convenience, but safeguarding your privacy and security should be a top priority. Understanding the data you share with extensions and their reliability in keeping it safe is crucial. By being cautious in choosing AI Chrome extensions and staying informed about their potential risks, users can embrace the benefits of AI while safeguarding their personal information,” advises Darius Belejevas, Head of Incogni.
The study analyzed 70 popular AI-powered extension extensions across 7 categories examining their risk metrics, permission requirements, and data collection practices. According to Incogni, the potential dangers associated with some AI extensions may be greater than you think.
Key Insights
- 69% of investigated extensions have a high-risk impact (if turned malicious, they could be highly damaging to users’ cybersecurity).
- 10 out of 10 AI-powered writing extensions have a high-risk impact (could do a lot of harm).
- 10 out of 70 analyzed extensions have a high-risk impact and a high-risk likelihood (are more likely to turn malicious).
- Over 59% of the investigated extensions collect user data, with 44% of them collecting personally identifiable information (PII).
Risk Impact and risk likelihood of the studied extensions
Risk impact is a measure based on the number of permissions requested by the extension. Extensions with a low-risk impact score can’t do much harm, even if they get into the wrong hands. Extensions with a high-risk impact, on the other hand, could be highly damaging if they get into the wrong hands because of the data they have access to.
Risk likelihood is related to the perceived probability of a Chrome extension turning malicious. According to Incogni, it is derived by considering the publisher’s and extension’s reputation on the Chrome Web Store, how long the extension has been available on the store, and other data points concerning the extension.
The majority of AI Chrome extensions analyzed have a high-risk impact (48 out of 70 have a high or very high-risk impact). However, 29 of those are low-risk likelihood.
All 10 AI-writing extensions analyzed were found to have a high-risk impact. In fact, this group alone accounted for over 20% of all high-risk impact extensions.
In addition, the research found 10 extensions that had both a high-risk impact and high-risk likelihood across various categories with the most common being personal assistants and audio-visual generators.
Permissions requested
The permissions requested by these extensions dictate how much access and control over your browser, system and data they have. The potential damage if certain permissions (for example the webRequest permission allows an extension to “observe and analyze traffic to intercept, block or modify requests in flight”) given to the extension get into the wrong hands could be too critical.
Each of the studied extensions requested 3.7 permissions on average with AI-powered personal assistants requesting up to 5. AI-powered coding assistants required fewer permissions on average (2.3). Some of the frequently requested permissions include tabs, storage and contextMenus.
Some permissions with a broad range of implications include:
- activeTab, required by 22/70 extensions which allow the extension access and control to the currently opened site. This permission can allow an extension to scan and manipulate the contents of the site.
- Scripting, which was required by 17/20 of the studied extensions(including Grammarly, Quilibot and ChatGPT). This allows the extension to inject JavaScript and CSS code into a user’s webpage. This can change the content and layout of the page.
- webRequest, required by 6/70 extensions. This extension can monitor and modify data sent between users and the websites they visit.
User Data collected
The AI-powered extensions studied collected an average of 1.4 data points each, including those that claimed they did not collect any data points at all.
Each of the seven categories contained at least one extension that collected Personally Identifiable Information (PII), user activity and website content. On the other hand, none of the extensions collected “health information”.
The researchers add, “Namely, it was observed that 31 out of 70 (44%) extensions collect personally identifiable information, which can include things like the user’s name, address, and identification number. 15 out of 70 (21%) extensions collect user activity, which can contain network monitoring and mouse position or keystroke logging. Lastly, 8 out of 70 (11%) extensions collect users’ locations.”
Overview of the 5 most popular AI Chrome extensions
Below are the top 5 AI-powered Chrome extensions:
- Grammarly is the most popular Chrome extension with 10 million users. The extension requests 5 permissions including Scripting with users giving up 5 data points including PII, user location and activity. Grammarly was found to have a high risk-impact with a very low-risk likelihood.
- QuilliBot is the second most popular Chrome extension with 2 million users. Users grant QuilliBot 7 permissions including activeTab and scripting while also giving up 4 data points including website content, PII, location and user activity. The extension was found to have a high-risk impact with a very low-risk likelihood.
- WebGPT: ChatGPT with internet access. The extension does not collect any data points and only asks for storage permission. WebGPT: ChatGPT with internet was found to have a high-risk impact and a very high-risk likelihood.
- Monica is the fourth most popular Chrome extension. The user gives up 4 data points including PII, personal communications, and financial and payment information. The extension also requests 3 permissions including scripting. Monica was found to have a high-risk impact with a low-risk likelihood.
- ChatGPT for Search – Support GP4. The extension does not collect any data points but asks for 4 permissions, including scripting. It has moderate risk impact and likelihood.
Despite the boost in performance and enhanced browsing experience AI-powered offered by AI-powered extensions, it is important to consider their associated risk metrics, user permissions required as well data collected.
