what i read this week - week 22 2026
Thought that after my post on summary distrust, I could share a list of what I read each week. I technically prefer to process and digest what I read into blog posts, but not everything makes it into one, and this is a way to document and keep them, and maybe give others some food for thought.
This is not necessarily stuff I fully agree with, I'm just sharing what ended up in my feed reader or was linked in stuff I read, and doesn't include all the personal blog posts I read.
I had a lot to catch up on because I read a lot less the previous 2 weeks.
online articles
AI detection was built for faces - article about how bad AI detection works for war and climate propaganda videos, as the detection mechanisms often rely on biometric human features, and cannot accurately detect fake fire, smoke effects, etc.
US Law Enforcement Warns of ‘Anti-Tech Extremism’ - the US gov is aware of the sentiment around AI and is willing to target and suppress it, and they have little paid accomplice firms too who keep surveilling you on social media and in real life meetings if you organize to oppose data centers or voice criticism about them.
Iran Israel AI war propaganda - The AI propaganda we see with armed conflicts right now is a dire warning to the future of online video information. Goes more in-depth about detection methods.
Can Tracking Private Jets Predict an Imminent Apocalypse? - article about a site that assumes the rich elites will find out about an apocalypse first and try to flee, therefore serving as a warning system to the rest of us.
Why GCC Nations Must Move Beyond Content Moderation to Regulate Harm by Design - GCC means Governments in the Gulf Cooperation Council. Article is about how certain countries have already heavily regulated (and, arguably, censored in their favor) social media platform content, so now they should do the same for platform design. Eh...
Big Tech Will Not Save Us From the Climate Crisis - Big Tech is moving away from their climate targets and carbon credit bullshit because they wanna do more AI and data centers. The rest they are doing is unproven or not working.
Definition of Overburdened Communities in New Jersey - data centers and other similar detrimental undertakings often target overburdened communities, and this is what it means.
A Town Hall Too Late - article documenting how citizens near an almost finished data center actually get informed and treated (not well). They only received information well after the thing started to get built. It is being developed by DataOne for the Nebius Group to support AI infrastructure as part of a $17 billion deal with Microsoft.
Meta loses High Court challenge - summary of the case and possible fine.
Responsible Innovation Harms Modeling on Microsoft's Learning Platform.
EU AI Omnibus Deal Changes - more analysis on the proposed AI Act changes, nudifier ban and more, prominent actors, Merz ruining everything for us as usual, etc.
The AI Act is not ready for agents - article for a paper that's also listed below; risks of agents, and a need for more guidance from the AI Office.
AI’s real threat is worker control and surveillance - about the divide between workers who use AI and those who are managed by it. Higher paid jobs can be supplemented and accelerated by it, while the less fortunate, less earning (warehouse, gig work) are suffering under AI micromanaging them, causing scheduling issues, errors and more, and are more intensely surveilled than ever by AI "bossware".
Entzauberung der Digitalen Souveränität - German; deconstructing the term "digital sovereignty" and ideas around it. Mostly about this talk.
AI Forensics gegen Big Tech - German; Interview with AI Forensics founder Marc Faddoul about his work and the fear of retribution, especially the fear about getting targeted by Elon Musk.
Human Rights Due Diligence - info on what downstream HRDD is.
Microsoft took a step towards human rights - very charitable and exaggerated read of Microsoft parting ways with their Israel chief and their ties to the Israeli Ministry of Defense, plus suspending some of their services.
The World Is Already Resisting AI - Article on the AI Resist List, a collaboratively built, publicly accessible database documenting acts of resistance to the AI industry from across the world.
AI Data Centers: Big Tech's Impact on Electric Bills, Water, and More - looking at different papers and studies around the water and electricity use of big data centers, where they are located, and what local problems they are worsening. Meta’s Hyperion project in Louisiana will need three times as much electricity as the entire city of New Orleans, and is bigger than its main airport. They also gag local officials with NDA's so they can't properly inform the residents.
What you need to know about data centers - information on what Earthjustice attorneys are doing to push for stronger environmental protections targeting data centers.
The Web Is Being Made Accessible for AI, Not People - llms.txt convention, MCP etc.; companies are more ready to make their services accessible to AI agents than disabled people. This shouldn't be seen as another curb cut phenomenon.
Bitte im Omnibus sitzen bleiben, liebe PIMS - German article about the Art. 88 reworks for Personal Information Management Systems that are supposed to enable an easier handling of cookie consent and tracking.
Social Media Verbot weder wissenschaftlich fundiert noch effektiv - German; about how there is no scientific proof that social media bans will help, and some stats about how many people support social media bans, and for what age group.
Big Tech und Staat - German article on how the state seems to increasingly serve private interests, especially Big Tech.
Bundesregierung will KI Einsatz der Polizei - German article about use of AI software for law enforcement, its risks, and what rights are threatened.
Polizeigesetznovelle Schleswig-Holstein - German article discussing Schleswig-Holsteins attempt at changing their police law, including real-time facial recognition, behavioral surveillance, online face search and more, from strangers on the street, and even mere victims or witnesses of crimes.
Das Internet verrottet - German; about link rot and archiving things properly.
Why “Made in Europe” Won’t Fix AI’s Deeper Problems - fitting to my blog post.
Big Tech as Executor of the dead - was also a topic at the conference.
press releases, policy briefs, drafts and activity reports
Praxisfolgen Russmedia Urteil - consequences for social media platforms following the Russmedia court decision C-492/23; Notice-And-Sweep.
AI Act: deal on simplification measures, ban on “nudifier” apps - concluding what deal was reached between co-legislators; names the new deadlines for AI compliance.
Ratepayer Protection Pledge by the White House - promises and propaganda
Microslop's Community-First AI Infrastructure Pledge - promises and propaganda vol. 2
Anthropic's Promises - promises and propaganda vol. 3
Offener Brief der Industrie - Open letter to German politicians by German industry criticizing parts of the digital omnibus; it was silly to read, and I think it is disrespectful to imply that technologies can be discriminated against; that's a different usage and connotation than just using it as "being discriminated from" (aka being differentiated from others). None of the arguments are convincing.
Draft guidelines for the implementation of transparency obligations for certain AI systems under Art. 50 AI Act - this is out for commenting until the 3rd of June, by the way.
Consent Fatigue entgegenwirken - German policy brief by the TUM think tank about countering consent fatigue.
papers, studies and similar publications
Einstellungen zum geplanten Einsatz von Palantir-Software II - German phone survey about Palantir use by Verian & campact from Sep 2025.
Grok Unleashed - Analyzing Grok nudify uses and extremist propaganda, by AI Forensics.
Distinguishing Authentic from AI-Generated Explosions using Spatiotemporal Dynamics - more about how to authenticate conflict-zone explosion footage. AI footage tends to produce much bigger, rounder mushroom plumes that expand quicker. Don't ask me about the math, I don't understand any of that, but I found the rest I could understand very interesting.
Embedding Human Rights in Technical Standards - About WITNESS' experience in the Coalition for Content Provenance and Authenticity (C2PA), which is in favor of open technical standards to embed verifiable provenance metadata into digital media files. Helpful explainer here.
Better Images of AI - a guide for creators and users on how to use accurate images when talking about AI and what to avoid, as it shapes the narrative. Specifically, they call to avoid the color blue, descending code, human brains, science fiction elements, white robots, anthromorphism and references to the Creation of Adam. That is because it misrepresents capabilities, risks and fears, and who is or can work in or with AI (often, only white men are shown).
The AI Climate Hoax: Behind the Curtain of How Big Tech Greenwashes Impacts - talks about how different kinds of AI and its uses as well as carbon credits and overstating the climate benefits of AI can be used to hide the environmental impact of the big, hyped up GenAI.
Big Tech’s ‘False Solutions’ to the Climate Crisis - similar thing here. Debunking nuclear power, carbon capture, and artificial intelligence as helping climate change. There are endnotes at each chapter, so don't miss what's after.
Tackling Arbitrary Digital Surveillance in the Americas - uses Cajar vs. Colombia for some examples to showcase what needs to change, and the importance of the three-step-analysis. Basically all of this is standard here in the EU, but still needs to be implemented there.
TRIED AI Detection Benchmark - paper from WITNESS about their framework that evaluates AI detection tools through a sociotechnical lens (with a focus on adaptability, transparency, accessibility, contextual relevance, and fairness). Wasn't a complete fan, because a chunk of it (for example about resource investments) is rather vague, theoretical and hardly connected with a direct or objective way to measure in practice. The rest is mostly fair, but also rather obvious, and some of it is basically impossible to combine in practice - like only using datasets that comply with data protection and intellectual property laws and are "ethical" with no sensitive data, while the models are supposed to reliably detect an AI generated video of a minority language or niche culture, or have enough datasets (= lots) to accurately detect cultural and local contexts. I can't quite pinpoint what exactly bothers me about it otherwise. I did like the examples of real use cases where things failed.
cases
- Zugänglichkeit von De-Personalisierungsoptionen und Meldeverfahren auf sehr großen Online-Plattformen
- Decisions I had to read to translate for noyb: 2025-0.875.804 and W171 2305420-1
In total, that is roughly ~ 340 pages, if we count an article as two pages on average. Most of it was read on Sunday and Monday (holiday), as I had a lot of free time then.
Reply via email
Published