Algorithms used to take down “terrorist” videos are deleting evidence of war crimes and could hinder efforts to bring human-rights abusers to justice
Algorithms that take down “terrorist” videos could hamstring efforts to bring human-rights abusers to justice. — Journalist based in Rome Tweets: @mbogen , @eff , @danteramos , @eff , and @jake_hanrahan Tweets: Miranda Bogen / @mbogen : This is fascinating: minutes from an internal Facebook meeting in December about redefining internal definitions of terrorism (shared by a company spokesperson with a journalist for @TheAtlantic) https://fbnewsroomus.files.wordpress.co m/ ... Via https://www.theatlantic.com/ ... @eff : If governments push social media companies to take down “terrorist content,” they will also force them to remove important documentation of everything from atrocities and war crimes to police brutality as well, say journalists and human rights activists. https://www.theatlantic.com/ ... Dante Ramos / @danteramos : Tech companies are using artificial intelligence to take down “extremist” content. But they're also robo-censoring evidence that could be used to thwart warlords, dictators, and other human-rights abusers. @BernhardWarner reports: https://www.theatlantic.com/ ... @eff : Social media is often the only source for documentation of war crimes and atrocities that journalists and human rights defenders desperately need to do their work. But new anti-terrorism bills will force platforms to censor them https://www.theatlantic.com/ ... Jake Hanrahan / @jake_hanrahan : This has been evident for ages and doesn't get any less disgusting. Good article. Have a read. https://www.theatlantic.com/ ...
Context & Ripple Effects
This story extends reporting from late 2018, when BuzzFeed documented how [[a:936996|proactive content removals by social networks were already hampering investigators and NGOs documenting war-zone atrocities]]. The Atlantic's piece adds the mechanism: the same classification algorithms built to catch ISIS propaganda cannot distinguish a beheading video used as prosecution evidence from one used as recruitment material.
The stakes are not hypothetical — by 2020, [[a:954715|dozens of Tunisian, Syrian, and Palestinian activists and journalists reported their Facebook accounts deactivated on terrorism pretexts]], showing the false-positive problem hitting the very documenters the platforms claim to protect.
First-order effects
- NGOs and independent investigators lose primary-source footage of atrocities as automated takedowns delete it before it can be preserved or submitted to courts.
- Activists and journalists in conflict regions face account deactivations under terrorism labels, cutting off their documentation channels entirely.
Second-order effects
- Human-rights organizations are pushed toward off-platform archiving workarounds, fragmenting the evidentiary record and raising costs for every future prosecution.
- Government pressure and new anti-terrorism bills compound the problem: each mandate to remove 'terrorist content' faster widens the deletion surface before human review can intervene.
Third-order effects
- If the pattern holds, accountability for war crimes increasingly depends on private moderation infrastructure that was never designed as an archive — creating a structural justice gap that regulators have yet to address.
- The asymmetry exposed here feeds broader demands, amplified after [[a:962094|the Trump suspensions prompted rights groups to urge Facebook and Twitter to apply policies evenly worldwide]], that enforcement systems distinguish perpetrators from documenters rather than defaulting to deletion.
The trend: Automated content moderation is colliding with human-rights documentation, turning platforms' counterterrorism tooling into an accidental destroyer of the evidentiary record for atrocity prosecutions.