I remember scrolling the Instagram account @eyeonpalestine for months, and how you start to feel like when you stay there long enough and become doomed by a sensation, becoming a fog. The feeling is both physical and emotional. It’s like being somewhere after an explosion. You got blurry vision, your ears are buzzing.
Some images are clear, unfiltered, and painful as they were meant to be: bodies, rubble, blood. Others are covered. Blurred. There is apparently no legible criteria, no consistent logic you can extract from what gets blurred and what does not. Extremely violent images sometimes appear in full while others, less graphic, less explicit, are swallowed by a grey screen, a pixelated veil, and a sign that resembles a blinded eye, an eye with an arrow upon it. The signs of what Instagram calls “sensitive content warning.”
From Exactly That Body: Images Against Oppression, Kateryna Iakovlenko, e-flux Journal, Issue #133
If you scroll long enough, it seems like you are exactly in the fog, part of the fog. Again, a doomed state where the blur starts to become the content itself: pixelled little squares and a never-ending blind eye staring back at you. A blind eye reminding you that, if you really want to, proof of atrocity can be accessed with just a click, a declaration of willingness. But let’s be clear. All of this can be accessed only if you really want to, only if you are brave, or voyeuristic enough to click.
Why would you want to, tho? Why are you still looking for destruction, atrocity, death? Why, scrolling on a feed through image after image of painful destruction, would you stop specifically on the blurred one and choose to unlock it? What are you trying to see?
The platform has already answered for you. We all know what you are looking for: proof, evidence, reasons. The platform is sorry, but all of these are really hard to find: evidence becomes too much to be allowed. Let’s keep it safe here: evidence is lost media; pixels are all you’re gonna see. Don’t worry though: we have evidence at home.
When something is too bad, there is always AI with a polished and less disturbing version of what you thought you were looking for. Flags, sad faces, refugee camps with orderly rows of tents, clean skies, no smoke. Remember All Eyes on Rafah, an AI-generated aerial image of a camp, shared 47 million times on Instagram in May 2024? The image’s creator acknowledged: “Real photographs of Gaza tend to have limited reach on Instagram, while the picture from AI can spread faster in a short time.”
Evidence that never existed in the first place is way safer. It’s just a matter of pixels put together or put apart and acting in different ways. Some are erasing, some are inventing–speculating. Both are making something else of the living. But before the AI fills the space, something else has to clear and make room for it. The blur has to come first.
But who is blurring? And why does the platform allow it?
In Meta’s policy language, the blur does not directly exist. If you search on meta.com for the word “blur”, what comes back is just information about Ray-Ban Meta, their smart glasses that the website is promoting right now. The website is a commodity, an augmented reality commodity. Instead, the grey screen, pixelated veil with an eye with an arrow upon it in the policy language is referred to as the ‘sensitive content warning’. Meta’s official position on graphic content is documented in their Transparency Center, under what is called a three-part enforcement approach: remove, reduce, inform.
- Content that violates Community Standards gets removed.
- Content that does not violate Community Standards but is considered ‘problematic’ in some way gets reduced, meaning that its distribution is limited
- In this case (2), users get informed with what Meta calls ‘additional context’: extra information so the user can independently ‘decide what to click, read or share.’
The blur, that sensitive content warning, lives in the reduce and inform category, a category Meta has maintained since 2016 as a grey zone between what is allowed and what is not. A category where content exists but cannot be easily found, where a body is technically present but functionally invisible. And most importantly, where the responsibility of the life of the content seems to be put completely on the user: it is you, scrolling, having the power to decide if that image exists or not.
For graphic content specifically, Meta is clear: “We do allow people to share some graphic content to raise awareness about current events and issues. In these cases, we may hide the content from people under 18 and cover it with a warning for those over 18, so people are aware it is graphic or violent before they choose to view it.”
In Meta’s words, the warning screen, the blur, is neither censorship nor moderation. It is just a design choice and a more-agency-to-the-user operation. A design choice written into policy, applied selectively to content Meta considers simultaneously worth keeping for someone and too much to show for someone else.
The ‘inform’ part of the three-part enforcement approach (remove, reduce, inform) claims that the warning screen exists to enable awareness for those who are looking for it—a helpful strategy to let graphic content circulate while protecting users from encountering it without consent. But the ‘inform’ doesn’t come without the ‘reduce’ part, and the helpful strategy produces the opposite effect: content behind a warning screen reaches fewer people, it circulates less, and the algorithm deprioritizes it. The awareness is technically there–just one click away–but the enforcement approach is built to make sure most users never make that click.
The closest thing Meta has to an internal court on these questions is the Oversight Board, an operational body that Meta itself created in 2020, ”funded through an independent trust, whose members are appointed to review the platform’s most controversial content moderation decisions”, as you can read on their page. Meta presented it as an “independent check on its own power,” and the Oversight Board’s official website explains that the Board’s role is to examine whether Meta’s decision to remove or leave up content was in line with its policies, values, and human rights commitments.
“While we refer to Meta’s content policies and values as we consider cases, we sometimes question those policies when we believe they do not comply with Meta’s own commitment to protecting freedom of expression and other human rights. When we come to a decision on a case, it is binding on Meta, unless implementing it could violate the law. As part of our decisions, we also make recommendations on the rules that apply to billions of Facebook, Instagram and Threads users, and how those rules are enforced.”
Moreover, in the ‘Governance’ section on the website, the Oversight Board claims that “Transparency, accountability and independence are central to the mission of the Oversight Board. Our governance structure allows the Board to reach independent decisions on matters of content moderation, which is critical to our success.”
Unsurprisingly, nowhere on the official website is it clearly declared that the Board was created by Meta itself in 2020. Nowhere is it stated that the Board can be seen as a washing PR system. It is claimed tho that “Recommendations also form part of our policy advisory opinions, which we started publishing in 2022 to provide guidance to Meta on specific issues.”
In 2022, Meta’s own Oversight Board examined the logic of the ‘sensitive content warning’ in a case involving a graphic video from Sudan documenting what was declared by the Board as “human rights abuses”. In that case, the Board found that Meta’s existing policy failed to“adequately protect content shared for documentation and accountability”. It recommended that Meta “amend its Violent and Graphic Content Community Standard to explicitly allow videos of people or dead bodies when shared for the purpose of raising awareness of human rights abuses”.
Apparently, this is where the final choice comes from: graphic and explicit content should be allowed when shared for the purpose of raising awareness of human rights abuses, but with a warning screen so that people are aware that content may be disturbing.
When I pitched this article to INC, the editor answered that “the images are not blurred by Meta itself, too much work. They delegate. Zuckerberg could not care less. Indeed, the act of blurring is a distributed system, layered and designed so that basically no direct actor can be held responsible for what disappears and gets moderated.
As Meta’s own Transparency Center states, generally the first level of moderation (1) is the users themselves. When other users report content, platforms flag it for review. Instead, the second level (2) is automated. Meta declares it uses AI and machine learning to detect and flag content before users even report it, based on a system trained (nowhere is explained how) “to identify graphic or sensitive material”. Needless to say, the system is biased. Especially when the ‘trained system’ is busy dealing with non-English text (and particularly Arabic texts). The systematic moderation of the word ‘shaheed’(martyr, in English) is just an example, and here we go again; even the Oversight Board quoted this example as a problem. To continue, the third level (3) can be the creators themselves. For example, Instagram allows users to pre-apply the sensitive content warning to their own posts. The idea is to blur your own content before the platform does it for you, hoping in this way it’s gonna be ‘reduced’ but not ‘removed’.
In charge of all of this is a small and unidentified group of workers. In 2024, Zuckerberg declared at the US Congress that Meta had approximately 40,000 people working on safety and security. When asked for details, he declined to answer written questions from the Senate Judiciary Committee about how many employees were specifically working on content moderation. This happened right after that, in 2023, Meta published what they proudly called the “Year of Efficiency”: an internal memo announcing the reduction of Meta’s workforce by approximately 10,000 people, with 5,000 additional open roles closed. One of the small pieces of information available, as reported by NBC News, is that “in January 2023, Meta’s primary content moderation subcontractor in Africa cut 200 employees, stepping back from content review entirely.”
Portrait Mode, from Portraiture, Surveillance, and the Continuity Aesthetic of Blur, Stefka Hristova, Frames Cinema Journal, 2021
In her essay ‘Portraiture, Surveillance, and the Continuity Aesthetic of Blur’ (Frames Cinema Journal, 2021), Stefka Hristova traces the genealogy of the blur and of the ways it has been used. Initially, because of the first technology for photography and technical limitations, people were photographed in sharp focus while backgrounds receded into a soft blur. Years later, when blur was no longer technically required, Apple shifted the language around blur to make it explicitly part of the photographic portrait aesthetic with its “portrait” mode.
What connects these moments, Hristova argues, is what she defines as the ‘aesthetics of continuity’: visual technologies that mask their function as control mechanisms by mimicking the appearance of more neutral, natural technology. As the ‘iPhone portrait mode’ blur imitates the depth-of-field effect of a camera lens, the grey screen on Instagram looks like a loading error, a technical glitch—something that’s harmless, designed for you rather than something that was decided.
The relation between “pretending to be natural” and “pretending to be for you” becomes perfectly understandable if we think that most images today are not made for human eyes but directly by machines and for machines. As argued by Trevor Paglen in Operational Images (e-flux Journal, Issue #59, 2014), images become operational rather than visual: they stopped being communication practices and have become infrastructure. In this way, “Instead of simply representing things in the world, the machines and their images were starting to ‘do’ things in the world.” And the blur is the only moment in which that infrastructure makes itself visible to the human eye, to protect itself.
The platform blurring, the ‘sensitive warning’, is the precise instant when the system pretends to remember the design choice of the platform is directed to the human on the other side and, for not being held accountable, asks them to take a position. But, as said before, let’s be clear: the presented choice is just a performance of choice, not a real choice. By the time the blur appears, the image has already been reduced. As explicit in the three-part enforcement approach (remove, reduce, inform), you are not choosing anything. You have been informed. Actually, someone just chose what is ‘sensitive’ for you.
What does “sensitive” actually mean and who decides this meaning? In a genocide, when does a body stop being political and become sensitive? Writing from Kyiv during the Russian invasion, Asya Bazdyrieva in ‘Exactly That Body: Images Against Oppression’ asks precisely this. “The notion of ‘sensitivity’ is always political. Who determines our exposure to ‘sensitive’ images? Who controls the mechanisms that distribute them? Is it tech companies, the state, or the individual who has survived the tragedy depicted in the images?” Cruel images and images of war are considered obscene and then blurred, accompanied by warnings, marked as sensitive. “But in reality, especially in the reality of war, ‘sensitive’ does not mean offensive.”
Drawing on artist Oraib Toukan’s work on cruel images, Bazdyrieva argues that, on the contrary, when words do not help, images can protect and defend. “A dead body does not cease to be a political body.” And so, let the image scream. Let the image be seen, even if it’s poor. Even if it’s hurting. Even if it’s sensitive. As argued by Hito Steyerl in ‘In Defense of the Poor Image’, “the poor image, pixelated, compressed, degraded by circulation, is not a failure but a political condition.” The pixel is the trace of movement: an image that has traveled, been copied, shared outside official channels, and survived despite its low resolution. “The poor image is a copy in motion; its quality is bad because it has been used.”
Thomas Ruff, jpeg rl104, 2007
However, the movement becomes a barrier on platforms. The blur on Instagram produces pixels too, but, contrary to a popular image pixelated because it has circulated too much, the content behind Meta’s warning screen is pixelated so that it will not circulate that much.
In The Perilous Potential of the Blur: Digital Cultures Within Zones of Indistinction (MAST Journal, 2023), Tony D. Sampson and Jernej Markelj describe the blur as a zone of indistinction, neither fully visible nor fully erased, neither present nor absent. A zone that implies “overlaps, collisions, interference, non-locations, vacillations, insensibility, inseparability, fuzziness, ambiguity, and even mess.” A zone not completely accessible and not completely remembered, functioning as a memory technology aiming to delay recognition, fragments recall, and turning past into something that can only be remembered through friction—a condition of seeing under occupation, exile, and historical erasure.
‘Your Father Was Born 100 Years Old and So Was the Nakba’, Razan AlSalah, 2017
In ‘Your Father Was Born 100 Years Old and So Was the Nakba’ (2017), a short movie by Razan AlSalah, Oum Ameen, a Palestinian grandmother, returns to her hometown Haifa through Google Maps Streetview, the only way she can still see Palestine. Here, Haifa only exists as a mediated memory, through an image system that is itself owned, partial, unstable, and obstructed. The longer Oum Ameen searches and tries to remember, the harder it is. The faster Oum Ameen tries to go, the slower the platform responds.
‘Your Father Was Born 100 Years Old and So Was the Nakba’, Razan AlSalah, 2017
Looking for Ameen, when she finally finds someone that could be him, Ameen is blurred. Visibility in blur is never transparent, never innocent, never whole. It is always compromised, possible only as a visual failure, as a broken interface between memory and territory. The image is compromised by distance, by displacement, by the technical and political conditions of seeing.
‘
Your ‘Your Father Was Born 100 Years Old and So Was the Nakba’, Razan AlSalah, 2017
The image is no longer a stable repository of facts, but a blurred space where memory is forgotten and, potentially, fabricated again. In ‘We can make rain, but no one came to ask’ (2004) by The Atlas Group, blur becomes a form of historical pressure. It marks the image as damaged, incomplete, or mediated. Still, in doing so it also works inside loss, using fragmentation, opacity, and technical interruption as ways to keep memory from being closed off by erasure.
We can make rain, but no one came to ask, The Atlas Group, 2004
In every case and in every application, blurring is an effect of infrastructural and systemic violence. In Disruptions, Taysir Batniji archives evidence of infrastructural inequality, arriving at the blur from external constraints. Displaced in Paris and unable to travel from Paris to Gaza, Batniji maintained contact with his mother and family through WhatsApp video calls constantly interrupted by poor connections. Network instability would dissolve faces, streets, and rooms into clusters of pixels, blurred lines, and solid blocks of color. A moderation not imposed directly by a platform’s content policy but produced by the infrastructure of power: who owns the internet they are trying to connect to? Meta owns the platform.
Disruptions, Taysir Batniji
The result of this process is to create a new aesthetic, a platform-friendly aesthetic: an aesthetic of absence, an absence-core, where the image is evoked while depoliticizing the scene, just with a reminiscence of the scene itself. In this way, the images allow empathy to emerge without engagement with the structural causes of suffering. The suffering is suggested, not shown; the cause is removed, and what remains is affect without accountability. And, in this process, the blur acts as the mechanism that prepares this aesthetic of absence, as an infrastructure of erasing. What is left behind the grey screen is a now- absent body, a pixel, a space that has been cleared and prepared for something else.
Because cleared spaces, in the logic of platforms, cannot stay empty: they have to get filled. If Palestinian photographic and video testimony, real images of bodies, destruction, and resistance, are consistently subjected to ‘reduction’, blurring, sensitive warnings, and algorithmic suppression, AI-generated imagery can circulate freely, algorithmically preferred because, as Meta explained, the platform removes content that is “particularly violent or graphic, such as videos depicting dismemberment, visible innards or charred bodies”. And AI content lacks the bodily specificity that exactly triggers content moderation systems.
In her upcoming book Speculative Violence, Donatella Della Ratta defines AI images as “something subtler, more insidious than fake news: harder to detect by algorithms, easier to accept as no apparent harm was involved, and perhaps far more pervasive and disruptive in its potential consequences.” The speculative violence is described as “an emerging politics of vision, in which power operates by capturing through imagination the not-yet to shape and orient beliefs and actions in the now.” And, obviously, Generative AI is central to this dynamic because of its unprecedented ability to give, in a seamless way, visual form to almost any imagined future, even the most violent and exclusionary, making it a particularly suitable device for rendering annihilatory and supremacist agendas.
Gaza as a beachfront resort, A Third Temple built where Al-Aqsa stood: visual prototypes preparing the ground for future policies, desires, and consensuses. “The synthetic visual regime emerging with generative AI is one in which plausibility, repetition, and virality determine legitimacy. In this new order, circulation itself becomes a form of validation, and visibility replaces verification. They do not seek to pass as real, nor do they operate within a logic of truth versus falsity. These images are speculative by nature, and so is the violence they suggest and enact. Their power lies not in deception, but in anticipation: in making a violent future imaginable, plausible, and increasingly acceptable before it ever occurs.”
“In a highly visual — indeed hyper-visual — culture such frames operate as performative blueprints: they prepare the imagination for what is to come”, Della Ratta continues.
Della Ratta’s starting point is Nick Land’s concept of hyperstition, defined as “effective culture that makes itself real”, a fiction that succeeds in projecting itself into reality. If a vision is seen enough times, by enough people, it can begin to function as social reality before it materially exists. “Projections that do not simply represent imagined futures but actively participate in bringing them into being.”
Here, the role of blur is not parallel to the speculative violence, but it works as its perfect infrastructure. Absence acts as a condition of new possibility, erasing the image and cleaning the space before it gets filled. The body has to disappear first. The testimony has to be made inaccessible. The warning screen has to be applied. The image has to be forgotten–or, at least, very hard to remember. As Della Ratta quoted, Arjun Appadurai writes that speculation operates in “the zone of the invisible”, the most profound location of which is the future. It’s in this context that the absence-the bad remembering-the ‘reduction’ of the image prepares the ground for the speculation to come. Again, as written by Donatella Della Ratta, “photorealistic surfaces of AI do not merely imagine a future Palestine; they visually construct the preconditions of that future, training the public to accept it as inevitable”.
Blur and speculation form, in this way, a sequence where the present is obscured so imagination is yet to come.
—
Geremia Trinchese is a PhD candidate in the Joint International Doctorate in Social Representations, Culture and Communication at Sapienza Università di Roma, the University Alexandru Ioan Cuza in Iași and the University of Pécs. His research critically examines algorithmic suppression and platform governance.