What is a deepfake?
Deepfake is a video, image, or audio generated by artificial intelligence that imitates the appearance and voice of a person, being so convincing at imitating reality that they can deceive both people and algorithms. “Deepfake” as a term comes from the combination of the words “deep learning” and “fake”, to represent some kind of fake content that is the result of deep learning technologies. In most cases, this is based on video manipulations where the software analyzes the source material and extracts part of it (usually the face and facial expressions), and then inserts and adapts it into another video. So-called face swaps or deep video portraits are the most common form of fakes. Deepfakes are used to induce confusion and manipulate the recipients of the files, either by making a politician say something in a video that they never actually said to affect them in a political campaign close to an election, or by including the image of a famous person (or anyone) in pornographic material with the aim of harming or blackmailing them. Although these deception techniques are still in their early stages, as the technology becomes more widespread and increasingly available, it will be deployed for more nefarious purposes and will become exponentially more difficult to unmask.
Why are deepfakes dangerous? Because they convincingly distort reality, deepfakes pose several dangers, such as online fraud, misinformation, deception, and revenge porn. The ability to impersonate a person in a live video or over the phone will pose a challenge for online verification systems, which rely on face or voice recognition, whether done by a human or automatically by software. According to Deepfake Report, 96% of the deepfakes in video format that they could identify online were pornographic videos. This might seem relatively harmless compared to financial fraud, but it is not if you consider that it can lead to extortion and blackmail, with the creator of the deepfake threatening to release a fake private video of someone, which seems so believable that the victim has no choice but to pay. Another problem highlighted by journalists and researchers, from the BBC to Wired and Forbes, is the potential for deepfakes to cause political and social harm through misinformation and false scandals, as the image of a public figure can be manipulated to show them doing or saying things they never did. For these latter reasons, deepfakes are considered a significant weapon in the arsenals of cyber warfare, where national governments seek to mitigate harmful deepfakes for the country, while creating the ability to use them against their own enemies, if necessary.
How to identify a Deepfake:
- What are the sources to look for to verify the authenticity of a video? It is always important to look for the original source of the recording, locating the person who first shared the deepfake video. Often, this helps to find out the context of the post and to check if the source material was more detailed after all. In this regard, social networks are especially vulnerable sources for the proliferation of deepfakes, as they prioritize content according to interaction metrics (i.e., how often it is shared and receives “likes”) rather than its accuracy or whether it was well-researched. This approach can allow cyber hoaxes, hyperbole, and misinformation to spread rapidly. Social media companies are considered platforms rather than publishers, meaning they do not have the same legal responsibilities as traditional news media. For this reason, social networks should never be prioritized over recognized news media outlets when it comes to checking the source and veracity of any kind of information.
- What is the importance of verifying the date and place the video was recorded? Most digital files (including videos and photos) have descriptive information inside known as “metadata”, which can be analyzed to verify the authenticity of the files, such as the date and time of creation and the geographical location. In the case of deepfakes, the metadata usually does not match the metadata of a real video or photo, exposing the fraudulent origin of the shared content.
- What technological tools can be used to verify the authenticity of a video? Advancements in AI have produced powerful tools to combat deepfakes: a) Sentinel: Sentinel, developed by Sensity, is a cloud-based solution that offers real-time deepfake detection with high accuracy. It leverages multiple techniques, including facial landmark analysis, temporal consistency checks, and blink detection, to identify inconsistencies in manipulated videos. b) Intel’s Real-Time Deepfake Detector (FakeCatcher): This tool, from Intel, focuses on speed and efficiency. Using Intel hardware and software, FakeCatcher analyzes subtle physiological details such as variations in blood flow in pixels, allowing for real-time detection with high accuracy (96%). c) WeVerify: This platform developed by WITNESS combines human expertise with AI analysis. Users can submit suspicious videos through the platform, where analysts review them alongside AI detection algorithms. This collaborative approach offers thorough assessment and broader context for deepfake identification. d) Microsoft’s Video Authentication Tool: This free tool from Microsoft analyzes videos and images, providing a confidence score indicating the likelihood of manipulation. It identifies inconsistencies in edge blending and subtle grayscale scale elements, often invisible to the human eye. e) Phoneme-Viseme Discrepancy Detector: This technique focuses on inconsistencies between spoken audio and lip movements in videos. It analyzes the alignment between phonemes (sound units) in speech and visemes (lip movements). Mismatches between these elements can indicate possible manipulation. f) Facia: Founded in 2021, Facia is a leading provider of facial recognition technology with a focus on security and identification. While not just a deepfake detection tool, Facia’s liveness detection solution boasts a 100% accuracy rate to thwart more than 53 different spoofing attempts, including deepfakes. This is achieved through advanced algorithms that can distinguish between a real person and a photograph, video, or mask.
- What visual or editing cues indicate that a video has been manipulated? a) Blinking: Humans typically blink once every 2-8 seconds, with each blink lasting between 1 and 4 tenths of a second. If you want to know if you’re dealing with a DeepFake, pay attention to the number of times the person blinks; you’ll notice they blink much less than a real person. The DeepFake algorithm cannot blink at the same speed as a human.
b) Face and neck: Does the body match the face, or does the posture match the facial expression? Most DeepFakes are primarily face swaps; changes in the body can only be implemented with great effort. If the body of the person to whom a face is applied has other features that do not match, such as tattoos, pronounced muscles, or a different skin color, it’s a sign of forgery.
c) Short clips: Although the technology is now very easy to use, the learning processes to create forgeries are still complex and laborious. Therefore, most DeepFakes that are shared are only a few seconds long. So if you have to verify a very short clip of unlikely content and there’s no obvious reason why the recording is so brief, it’s often because it’s a fake.
d) Sound to picture: Not only does the image expose a DeepFake, but so does the sound. The software often limits itself to changing the image but not adjusting the sound. So if there is no sound or if it doesn’t match the image, for example, with poorly implemented lip-sync, this indicates forgery.
e) Details: Pay attention to details. When verifying the video content, it’s also helpful to let it play at half-speed. For example, small discrepancies in the background of a person or sudden changes in the image that are noticed immediately.
f) Inside the mouth: The software to create DeepFakes has been able to transfer faces quite well so far, but the devil is in the details. For example, a certain blur inside the mouth is another sign that it could be a fake image, as Artificial Intelligence is still struggling to correctly represent teeth, tongue, and oral cavity when speaking.
- What steps should a person take before sharing a video on social media to ensure it’s not a fake video? Fake videos rely on believers to repost, tweet, or share false information. If you’re not sure if a multimedia file is authentic or not, pause for a moment and think before sharing. It is important to first check the source and author, investigating to see if they are credible or have a good reputation. It’s also a good practice to check other sources to verify if the same information has been shared through other reputable news media outlets.
You can use a fact-checking site like: Snopes PolitiFact Fact Check BBC Reality Check
- Tips for recognizing a fake video of fake news a) Check the source: Verify the web address of the page you are visiting. Sometimes, fake news sites often have spelling errors in the URL or use unconventional domain extensions such as “.infonet” or “.offer”. If you are not familiar with the site, take a look at the “About us” section.
b) Check the author: Investigate to see if they are credible: for example, are they real? Do they have a good reputation? Do they write about their specific area of expertise? Do they have any particular intention? Consider what the writer’s motivation might be.
c) Check other sources: Are there other reputable news or information outlets covering the story? Are there credible sources cited within the story? Global professional news agencies have editorial standards and extensive resources to verify facts, so it’s a good sign if they are also covering the story.
d) Maintain a critical mindset: Many fake news stories are cleverly written to provoke strong emotional reactions such as fear or anger. Maintain a critical mindset and ask yourself, for example, why was the story written? Does it promote a particular agenda or cause? Does it try to get you to click to access another website?
e) Check the facts: Credible news will include many facts: data, statistics, quotes from experts, and more. If any of that is missing, ask yourself why. Reports with false information often have incorrect or altered dates, so it’s a good idea to see when the article was published. Are they current or old news?
f) Check the comments: Even if the article or video is legitimate, the comments below may not be. Often links or comments posted in response to the content are autogenerated by bots or people hired to place misleading or confusing information.
g) Check if it’s a joke: Satire websites are popular, and sometimes it’s not clear if a story is just a joke or parody. Check the website to see if it’s known for its satire or for creating funny stories.
h) Check if the images are authentic: The images you see on social media could be edited or manipulated. Some possible signs include distortion (straight lines in the background that look wavy), as well as strange shadows, cropped edges, or overly perfect skin tones. Also, keep in mind that an image could be accurate, but it was simply used in a misleading context. You can use tools like Google’s reverse image search to verify where an image comes from and if it was altered.
i) Use a fact-checking site: The following are some of the most recognized ones: o Snopes o PolitiFact o Fact Check o BBC Reality Check
- What precautions should be taken when watching a video on social media to avoid falling for misinformation? It’s important to note that anyone with a social media account can become a content creator. For individuals and media that publish content, the goal is almost always to be seen by as many people as possible. However, what proliferates on social media is not exactly balanced and well-researched stories. For posts on large platforms, “outrage is the key to virality.” In fact, researchers at MIT have found that fake news can spread up to 10 times faster than real reports on social media. In these cases, when explosive and poorly informed posts go viral, instances where they are corrected or refuted are never as seen or believed. The scandalous “fact” that bursts through audiences is stronger, stickier, and more interesting than a truthful but less sensational correction. In the race between the fake but interesting and the true but boring story, the interesting story always wins, so we must be cautious about this type of video and “extraordinary” content to avoid falling for misinformation.
- Why is it important not to share a video without verifying its accuracy first? The Internet created new ways to publish, share, and consume content and information, with relatively few normative or editorial standards. Many people now consume videos, articles, and news from social networks.
Document prepared in partnership
LIBEDINSKY CYBERSECURITY SERVICES
Fuentes
https://seon.io/es/recursos/glosario/deepfake/
https://www.iebschool.com/blog/deepfakes-como-detectarlas-business-tech/
https://www.lisainstitute.com/blogs/blog/deepfakes-tipos-consejos-riesgos-amenazas
https://seon.io/es/recursos/glosario/deepfake/#h-por-qu-son-peligrosos-los-deepfakes
https://www.iebschool.com/blog/deepfakes-como-detectarlas-business-tech/
https://latam.kaspersky.com/resource-center/preemptive-safety/how-to-identify-fake-news
https://www.linkedin.com/pulse/5-best-ai-deepfake-detector-tools-2024-machine-learning-1-vnxbf/?trk=organization_guest_main-feed-card_feed-article-content
https://www.iebschool.com/blog/deepfakes-como-detectarlas-business-tech/
https://latam.kaspersky.com/resource-center/preemptive-safety/how-to-identify-fake-news
https://latam.kaspersky.com/resource-center/preemptive-safety/how-to-identify-fake-news
https://pirg.org/edfund/articles/misinformation-on-social-media/
https://latam.kaspersky.com/resource-center/preemptive-safety/how-to-identify-fake-news
