Artificial intelligence (AI) is making its way from high-tech labs and Hollywood plots into the hands of the general population. ChatGPT, the text generation tool, hardly needs an introduction and AI art generators (like Midjourney and DALL-E) are hot on its heels in popularity. Inputting nonsensical prompts and receiving ridiculous art clips in return is a fun way to spend an afternoon.
However, while you’re using AI art generators for a laugh, cybercriminals are using the technology to trick people into believing sensationalist fake news, catfish dating profiles, and damaging impersonations. Sophisticated AI-generated art can be difficult to spot, but here are a few signs that you may be viewing a dubious image or engaging with a criminal behind an AI-generated profile.
To better understand the cyberthreats posed by each, here are some quick definitions:
AI art and deepfake aren’t technologies found on the dark web. Anyone can download an AI art or deepfake app, such as FaceStealer and Fleeceware. Because the technology isn’t illegal and it has many innocent uses, it’s difficult to regulate.
How Do People Use AI Art Maliciously?
It’s perfectly innocent to use AI art to create a cover photo for your social media profile or to pair it with a blog post. However, it’s best to be transparent with your audience and include a disclaimer or caption saying that it’s not original artwork. AI art turns malicious when people use images to intentionally trick others and gain financially from the trickery.
Catfish may use deepfake profile pictures and videos to convince their targets that they’re genuinely looking for love. Revealing their real face and identity could put a criminal catfish at risk of discovery, so they either use someone else’s pictures or deepfake an entire library of pictures.
Fake news propagators may also enlist the help of AI art or a deepfake to add “credibility” to their conspiracy theories. When they pair their sensationalist headlines with a photo that, at quick glance, proves its legitimacy, people may be more likely to share and spread the story. Fake news is damaging to society because of the extreme negative emotions they can generate in huge crowds. The resulting hysteria or outrage can lead to violence in some cases.
Finally, some criminals may use deepfake to trick face ID and gain entry to sensitive online accounts. To prevent someone from deepfaking their way into your accounts, protect your accounts with multifactor authentication. That means that more than one method of identification is necessary to open the account. These methods can be one-time codes sent to your cellphone, passwords, answers to security questions, or fingerprint ID in addition to face ID.
Before you start an online relationship or share an apparent news story on social media, scrutinize images using these three tips to pick out malicious AI-generated art and deepfake.
Fake images usually don’t appear by themselves. There’s often text or a larger article around them. Inspect the text for typos, poor grammar, and overall poor composition. Phishers are notorious for their poor writing skills. AI-generated text is more difficult to detect because its grammar and spelling are often correct; however, the sentences may seem choppy.
Does the image seem too bizarre to be real? Too good to be true? Extend this generation’s rule of thumb of “Don’t believe everything you read on the internet” to include “Don’t believe everything you see on the internet.” If a fake news story is claiming to be real, search for the headline elsewhere. If it’s truly noteworthy, at least one other site will report on the event.
AI technology often generates a finger or two too many on hands, and a deepfake creates eyes that may have a soulless or dead look to them. Also, there may be shadows in places where they wouldn’t be natural, and the skin tone may look uneven. In deepfaked videos, the voice and facial expressions may not exactly line up, making the subject look robotic and stiff.
Fake images are tough to spot, and they’ll likely get more realistic the more the technology improves. Awareness of emerging AI threats better prepares you to take control of your online life. There are quizzes online that compare deepfake and AI art with genuine people and artworks created by humans. When you have a spare ten minutes, consider taking a quiz and recognizing your mistakes to identify malicious fake art in the future.
To give you more confidence in the security of your online life, partner with McAfee. McAfee+ Ultimate is the all-in-one privacy, identity, and device security service. Protect up to six members of your family with the family plan, and receive up to $2 million in identity theft coverage. Partner with McAfee to stop any threats that sneak under your watchful eye.
The post How to Spot Fake Art and Deepfakes appeared first on McAfee Blog.
Authored by Fernando Ruiz
The popularity of AI-based mobile applications that can create artistic images based on pictures, such as the “Magic Avatars” from Lensa, and the OpenAI service DALL-E 2 that generates them from text, have increased the mainstream interest of these tools. Users should be aware of those seeking to take advantage to distribute Potential Unwanted Programs (PUPs) or malware, such as through deceptive applications that promise the same or similar advanced features but are just basic image editors or otherwise repackaged apps that can drain your data plan and battery life with Clicker and HiddenAds behaviors, subscribe you to expensive services that provide little or no value over alternatives (Fleeceware), or even steal your social media account credentials (FaceStealer).
Dozens of apps surface daily claiming to offer AI image creation. Some of them might be legitimate or based on open-source projects such as Stable Diffusion, but in the search for a free application that produces quality results, users might try new apps that could compromise their privacy, user experience, wallet and/or security.
The McAfee Mobile Research Team recently discovered a series of repackaged image editors on the Google Play app store which presented concerning behaviors. McAfee Mobile Security products help protect against such apps, including those classified as Android/FakeApp, Android/FaceStealer, Android/PUP.Repacked and Android/PUP.GenericAdware.
McAfee, a member of the App Defense Alliance focused on protecting users by preventing threats from reaching their devices and improving app quality across the ecosystem, reported the discovered apps to Google, which took prompt action and the apps are no longer available on Google Play.
We now discuss various types of privacy and/or security risks associated with the types of apps recently removed from the app store.
“Pista – Cartoon Photo Effect” and “NewProfilePicture” are examples of apps that offered compelling visual results, however, each was based on the same image editor with basic filters and trojanized with Android/FaceStealer, which is a well-known malware capable of compromising a victim’s Facebook or Instagram account. The apps could capture user credentials during a Facebook login by embedding a javascript function loaded from a remote server (to evade detection) into a flutter webview activity that displays the Facebook login screen.
“NewProfilePicture” and “Pista – Cartoon Photo Effect” are examples of FaceStealer malware that posed as a cartoon avatar creator.
The same image editor which was repackaged into the above two apps has also been repackaged alternatively with adware modules and distributed by other developers under other app names, such as “Cartoon Effect | Cartoon Photo”:
Fleeceware refers to mobile apps that use various tactics to enroll users into subscriptions with high fees, typically after a free trial period, and often with little or no value to the subscriber beyond cheaper or free alternatives. If the user does not take care to cancel their subscription, they continue to be charged even after deleting the app.
“Toonify Me”, which is no longer available on the Play Store, cost $49.99 per week after 3 days – almost $2,600 per year – for what featured AI-generated illustrations in the app description, but was another repackaged version of the same image editor functionality found within “NewProfilePicture” and “Pista – Cartoon Photo Effect”.
In this case, the “Toonify Me” app did not allow feature access without enrolling in the subscription, and the “CONTINUE” button which initiated the subscription was the only option to tap in the app once it was launched.
Promoted by ads that described it as capable of transforming pictures into artistic drawings, the “Fun Coloring – Paint by Number” app is an example of a repackaged version of a different, legitimate pixel painting app. It lacked the advertised AI effects and was plagued with adware-like behavior.
Advertisement of “Fun Coloring – Paint by Number” on social media which included app store link
Consistent with many reviews complaining about unexpected ads out of the context of the app, once installed, the app started a service that communicated in the background with Facebook Graph API every 5 seconds and might pull ads based on received commands after some time of execution. The app contained multiple injected SDK modules from AppsFlyer, Fyber, InMobi, IAB, Mintegral, PubNative and Smaato (none of which are in the original app, which was repackaged to include these), which would help monetize installations without regard for user experience.
When new types of apps become popular and new ones appear on the market to offer similar features, users should act with caution to avoid becoming victim to those wanting to exploit public interest.
When installing an app that causes you doubt, make sure you:
Even if an app is legitimate, we also encourage users to look closely before installation at any available privacy policy to understand how personal data will be treated. Your face is a biometric identifier that’s not easy to change, and multiple pictures might be needed (and stored) to create your model.
Artificial intelligence tools will continue to amaze us with their capabilities and probably will become more accessible and safer to use over time. For now, keep in mind that AI technology is still limited and experimental, and can be expensive to use – always consider any hidden costs. AI also will bring more challenges as we discussed on the 2023 McAfee Threat Prediction blog.
The following table lists the application package name, hash sum SHA256, the minimum number of installations on Google Play, and the type of detected threat. These apps were removed from Google Play, but some may remain available elsewhere.
Package Name | SHA256 | Installs | Type |
com.ayogamez.sketchcartoon | 9cb1d996643fbec26bb9878939735221dfbf639075ceea3abdb94e0982c494c1 | 5M | Adware |
com.rocketboosterapps.toonifyme | 3f45a38b103e1812146df8ce179182f54c4a0191e19560fcbd77240cbc39886b | 10K | Fleeceware |
com.nhatanhstudio.cartoon.photoeffect | 2c7f4fc403d1449b70218624d8a409497bf4694493c7f4c06cd8ccecff21799a | 5K | Repackaged Adware |
com.cambe.PhotoCartoon | 5327f415d0e9b21523f64403ec231e1fd0279c48b41f023160cd1d70dd733dbf | 10K | Repackaged Adware |
com.chiroh.cartoon.prismaeffect | 18fef9f92639e31dd6566854feb30e1e4333b971b05ae9aba93ac0aa395c955b | 1K | Repackaged Adware |
cartoon.photo.effect.editor.cartoon.maker.online. caricature.appanime.convert.photo.intocartoon |
3b941b7005572760b95239d73b8a8bbfdb81d26d405941171328daa8f3c01183 | 50 | Repackaged Adware |
com.waxwell.saunders.pistaphotoeditor | 489d4aaec3bc694ddd124ab8b4f0b7621a51aad13598fd39cd5c3d2067b950e5 | 50 | FaceStealer |
com.ashtoon.tooncool.skordoi | 980c090c01bef890ef74bd93e181d67a5c6cd1b091573eaaf2e1988756aacd50 | 100K | FaceStealer |
com.faceart.savetoon.cartoonedit | 55ffc2e392280e8967de0857b02946094268588209963c6146dad01ae537daca | 100 | FaceStealer |
com.okenyo.creatkartoon.studio | e696d7304e5f56d7125dd54c853ff35a394a4175fcaf7785d332404e161d6deb | 500K | FaceStealer |
com.onlansuyanto.editor.bading | 59f9630c2ebe4896f585ec7722c43bb54c926e3e915dcfa4ff807bea444dc07b | 10K | FaceStealer |
com.madtoon.aicartoon.kiroah | c29adfade300dde5e9c31b23d35a6792ed4a7ad8394d37b69b5cecc931a7ad9f | 100K | FaceStealer |
com.acetoon.studio.facephoto | 24cf7fcaefe98bc9db34f551d11906d3f1349a5b60adf5fa37f15a872b61ee95 | 100K | FaceStealer |
com.funcolornext.beautyfungoodcolor | b2cfa8b2eccecdcb06293512df0db463850704383f920e5782ee6c5347edc6f5 | 100K | Repackaged Adware |
The post The Rise and Risks of AI Art Apps appeared first on McAfee Blog.