a review of the policies of Facebook, Instagram, Twitter and LinkedIn

Deepfakes pose a challenge for social platforms. This type of manipulated content, consisting of editing a video, audio or image using artificial intelligence to show, for example, a person of public relevance saying something that they have not really said, is a challenge for some platforms in which viral content tends to spread rapidly and, depending on the contexts, it can lead to misunderstandings.

Thus, and after the recent Facebook announcement about deepfakes, we have asked ourselves: What policies do the main social platforms have regarding content manipulated with artificial intelligence? To clear up any doubts, we have reviewed the Facebook, Instagram, LinkedIn and Twitter conditions and, in addition, reviewed what actions other actors in the technological world are taking to help combat them.


We start with the most popular social network in the West: Facebook. Monika Bickert, Facebook’s vice president of global policy management, explained on the company’s corporate blog that “there are people who get involved in manipulating the media to deceive” and that, “although these videos are still rare on the Internet” , present a “significant challenge for our industry and society as its use increases. “

From the company they have explained that, after having consulted “more than 50 experts worldwide”, they have opted for a simple policy: eliminate counterfeit content as long as they are complied with. these two requirements:

  • Has been edited or synthesized, beyond the clarity and clarity settings, in a way that is not clear to the average person and is likely to mislead you into thinking that the protagonist of the video said words that he did not actually say.

  • Yes it is product of artificial intelligence or machine learning, merging, replacing or overlaying content in a video to make it appear authentic. “

Now there are exceptions. Videos manipulated for satirical and parodic purposes and those that have been edited to omit words or change their order will not be deleted. That is to say, that, in one way or another, it is still possible to publish manipulated content. There is another caveat, and that is content posted by politicians will not be censored, even if they post deepfakes about their rivals.

Parody, satirical videos and those published by politicians are exempt from the application of the criteria mentioned above.

The reason? The newsworthy. According to Andrea Vallone, spokesperson for Facebook, the platform applies “the same standards as other users when it comes to manipulated media” although “if the content is newsworthy, even if it violates the rule, we will evaluate case by case whether it should be allowed to remain. on the platform “. Mashable asked Vallone what would happen if newsworthy content were to be classified as manipulated (that is, that it met the two criteria above) and there was no response. The clearest example of this exception is Nancy Pelosi’s video.

Finally, from Facebook they clarify that “in accordance with our existing policies, audio, photos or videos, whether false or not, will be removed from Facebook if they violate any of our other community rules, including those governing nudity. , graphic violence, voter suppression and hate speech. “


In her post, Monika Bickert refers only and exclusively to Facebook. Instagram is not mentioned in any way since, in fact, they have different community standards even though they are from the same company, FACEBOOK, in capital letters. Instagram’s policy against deepfakes is as simple as there is no policy against deepfakes, at least for now.

We surely remember how in June 2019 a deepfake by Mark Zuckerberg appeared on Instagram and how the platform chose to leave it published (Although CBS asked Facebook to remove it because the video contained its trademark). At the time, a company spokesperson told The Verge that “we will treat this content the same way we treat all misinformation on Instagram” and that “if third-party fact-checkers mark it as false, we will filter it from the Instagram recommendation surfaces like Explore and hashtag pages. ” The same policy as fake news.

Shortly after, Adam Mosseri, CEO of Instagram, gave an interview to CBS and explained that “we do not currently have a policy against deepfakes“and that” we are trying to assess whether we wanted to do that and, if so, how we would define deepfakes. “According to Mosseri:

“We need to have defined principles and we must be transparent about those principles […] Right now, we are trying to balance security and dialogue, and that balance is very, very complicated. Right now, the most important thing to focus on is getting to content faster. Once we can do that, we can discuss whether to remove it when we find it. “

Last December, Instagram explained how it will combat false information on Instagram and, incidentally, announced the global expansion of the fact-checking program. The policy is similar to Facebook, insofar as “if an external fact checker has rated content as false or partially false, we will reduce its distribution; we will remove it from the “Explore” section and hashtag pages. In addition, it will be tagged so that people can decide for themselves what to read and share, and what to trust. If these tags are applied, they will be shown to everyone viewing the content around the world, in news, profile, stories, and direct messages. “


In the line of Instagram is Twitter. At the moment it does not have a policy regarding deepfakes, but is asking users for their opinion to “shape our approach to synthetic and manipulated content”. This, according to Del Harvey, VP of Security and Trust at Twitter, responds to that they want us to its rules “reflect the voice of the people who use Twitter”.

Twitter does not have an exact definition for manipulated or synthetic content, but the first approximation is the following:

“We propose to define synthetic and manipulated media as any photo, audio or video that has been significantly altered or manufactured in a way that is intended to mislead people or change its original meaning.”

Harvey also exposes “a draft” of what they will do when detecting synthetic and manipulated media that “intentionally try to mislead or confuse people.” According to Harvey, Twitter can:

  • “Post a notice next to Tweets that share synthetic or manipulated media.
  • Warn people before sharing or liking Tweets with synthetic or manipulated means.
  • Add a link, for example, to a news article or a Twitter Moment, so that people can read more about why various sources believe that the media is synthetic or manipulated. “

It further stresses that “if a Tweet that includes synthetic or manipulated media is misleading and could threaten someone’s physical safety or cause other serious harm, we can remove it.” However, this is a draft and for the moment there is no firm policy. Users were able to give their opinion until November 27, 2019 and from Twitter they said they will make “an announcement at least 30 days before the policy takes effect.”


We come to LinkedIn, which has not been saved from deepfakes. The most notorious case is that of Katie jones, a fake profile that connected with people in the Washington DC political sphere and that came to light after the Associated Press discovered some flaws in the profile photo that indicated that it had been generated with AI (we already know that this is more than possible). Some have called this profile a “spy” and, according to William Evanina, director of the US National Center for Counterintelligence and Security, “foreign spies routinely use fake social media profiles to target American targets.”

LinkedIn does not have a policy against deepfakes, although yes against false profiles, which are strictly prohibited. It has an “automatic defense” system based on artificial intelligence and machine learning that during the first half of 2019 knocked down 98% of the false accounts that were created. In total, in the first half of 2019, actions were taken against 21.6 million fake accounts.

There are also publication rules in which it is remembered “to be professional and not to publish anything deceptive, fraudulent, obscene, threatening, hateful, defamatory, discriminatory or illegal”, without defining exactly what is “misleading” and “fraudulent”. Linkedin may “restrict, suspend or terminate your LinkedIn account and / or disable your articles for any violation of the ‘User Agreement’” and disable “accounts found using infringing content”, but no explicit mention is made of deepfakes.

What tech platforms are doing to help fight them

Given the policies regarding deepfakes on platforms, it is appropriate to take a look at the technology sector to see what are companies like Google, Facebook or Adobe doing to aid in deepfake detection.

Facebook, for example, claims to be able to detect a deepfake on its platform if it has previously had the opportunity to analyze the original file, although this does not always have to be the case. To develop detection tools, Facebook (in collaboration with Microsoft and other companies) has created a deepfake dataset and the Deepfake Detection Challenge to encourage developers. Both initiatives are subsidized with 10 million dollars.

Something similar is doing Google, which in collaboration with Jigsaw, has released a huge dataset of deepfakes that has been incorporated into the Technical University of Munich and the Federico II University of Naples and that can be downloaded from GitHub. This data set seeks to offer a sort of training basis for detection algorithms and has already been downloaded by more than 150 research and industry organizations.

Finally, it is worth highlighting the case of Adobe. The Photoshop developer company has a tool that, according to Adobe, is capable of detecting when the face of a photo has been manipulated with its tool. It is possible thanks to a convolutional neural network that in the first tests obtained 99% accuracy.

Even so, whenever there is a solution, a system capable of violating it is created. Ragavan Thurairatnam, co-founder of Dessa, put it quite well: it will be “inevitable that malicious actors will move much faster than those who want to stop them.”