When Algorithms amplify Tragedy: The Kirk case and Social media accountability
Preventing the algorithmic amplification of traumatic content: we need more safe scrolls.
Snapshot 👀 TL;DR
On September 10, 2025, conservative political TikToker Charlie Kirk was fatally shot during an event at Utah Valley University. Within hours, graphic videos showing him bleeding from a gunshot wound were circulating across major platforms: his final moments amplified to millions by social media algorithms.
How much responsibility should platforms bear when their algorithms transform someone's final moments into viral content?
Prevent / Comply / Educate:
how to prevent the algorithmic amplification on social media of traumatic content,
how to align platform practices with human dignity protections set by regulators, and
how to increase a culture of ethical sharing online.
When Technology Lost Its Humanity
I saw the announcement and a warning on Linkedin. Thankfully, it was not the video on autoplayed in my feed. But imagine, some youngsters receiving in their own feed a traumatic video of someone's final moments: no warning, no choice. An algorithm couldn't distinguish between engaging personalised content and human tragedy.
Charlie Kirk’s death, a political content creator, was transformed into viral content by the same recommendation systems that had made him famous.
What’s crazy was how quickly the algorithms distributed the viral content. TikTok, Twitter, Instagram, and YouTube are all part of this problem. The same AI that optimizes for engagement couldn't safeguard its users against seeing traumatic death.
This isn't just failure of content moderation: it shows how certain values aren’t part of the system.
⚖️ EU Regulatory Reality Check
Current EU frameworks are regulating online content, but they are slightly fragmented:
Digital Services Act : While the DSA requires platforms to assess risks and implement mitigation measures, the platform responses can vary drastically. YouTube removed some content and Meta can either mark the content as sensitive or reduce its distribution. We’ve also seen how many platforms are taking a less restrictive approach regarding violent content. The autoplay function available from scrolling in a mass of content you don’t follow might bring more harm than benefits.
EU AI Act: Recommendation algorithms qualify as AI systems requiring transparency obligation, but this remains insufficient as recommendation algorithms always increase the risk of echo chamber. When algorithms maximize engagement without filters, there is a risk that systems are designed to exploit trauma responses and cause harm. We could also think whether those algorithms could qualify as exploiting vulnerability, especially for minors receiving directly in their feed those traumatic video content.
GDPR and Charter of Fundamental Rights of the European Union : protect fundamental rights but platforms argue they're not required to remove violent content unless explicitly illegal. This creates an accountability paradox: algorithms boost traumatic content for engagement while human dignity of the victim(s) is discarded as their traumatic incident is served to millions online.
💡 Moving forward: Dignity-by-Design
It is important to implement safeguard throughout the whole product lifecycle. We explore what could be good starting points:
Prevent:
Implementing better content filters that automatically detect graphic violence involving identifiable victims before amplification. Not only to individual users but to the whole platform.
Building recommendation systems trained on dignity preservation, not just engagement metrics. Age should also be taken more seriously into consideration.
Comply:
Having an efficient takedown system for viral content.
Implementing mandatory impact assessments for new algorithmic features and specially for content filtering.
Educate:
Implementing better warning disclaimer before the user accepts to see a video, potentially adapted to the user’s age.
Creating features that encourages users to consider human impact before amplifying harmful content: if the user uploading is conscious of the potential harmful impact or the viewer could also rank the content (1-10 potential harm/potentially shocking).
As algorithms become more sophisticated at distributing engaging and viral content, we must ask: should recommendation systems prioritize human dignity over engagement metrics? How much responsibility should platforms bear when their algorithms transform someone's final moments into viral content?
🧬 About Purple Mirror
Purple Mirror is your lens on what the future could look like—if disruptive technologies were designed with human values in mind.
Think of Black Mirror with a twist
I publish two format
🔮 Beyond the Glitch
Under 1000 words newsletter that explore emerging technologies at the intersection of innovation, ethics, and compliance.
⚡ Daily Glitch
A bite-sized ethical scenario
Each Glitch is a mini thought experiment—a weird, thrilling, or uncomfortable “what-if” from the near future.
One minute of reading. One vote. One step into tomorrow.
✍️ I’m Nelsy, sharing my personal insights on Law, Ethics, and Disruptive Tech.
📢 The views expressed here are solely my own. Any content published on this platform is intended to spark discussion and does not constitute legal advice.