what i read this week - week 24 2026
As usual - not counting the personal blogs I read :)
online articles
Not much appealed to me this week.
The AI ‘Revolution' is Not a People's Revolution - AI companies overusing the term revolution is just a marketing ploy, and we should challenge it. Banger quote: "Accepting Blair’s revolution requires agreeing that using unconsented data harvested from populations, processed through biased algorithms and presented to people in addictive interfaces that overwhelmingly generate wealth to US elites, is the change that people want."
Trump Signs Previously Shelved AI Executive Order- summary of the EO.
Widerstand gegen Kameras - German article about resistance against surveillance cameras; its history, methods and legal consequences. Person in the comments has an interesting tip: A brush, and acrylic paint mixed with sand.
UN-Report zu KI-Umweltkosten - German article about how the UN had the chance of holding tech companies accountable in a new report, but instead only asks consumers to adjust their behavior. I am not opposed to also asking people to rethink their consumer decisions (otherwise, I would not resist using animal products, flying, getting a driver's license etc.; if there's no buyer, there's no product), but for the biggest impact, we need to focus on the source and hold companies to a high standard - or ban their business model or product entirely. The report was also seen as low quality by experts in the field(s).
press releases, policy briefs, drafts and activity reports
Appeals Centre Europe Transparency Report April 2025 - March 2026 - The Appeals Centre is an independent out-of-court dispute settlement body active due to the Digital Services Act; they've only been around for 18 months. If you are in the EU, you can use them to challenge social media platforms’ decisions on groups, pages accounts or other content which has been removed or kept up despite reporting (if it is about anything other than impersonation, hacked accounts, copyright or CSAM, but hopefully those too at some point). Most cases seem to be about account suspensions, nudity, fraud and scams. So far, they have processed more than 24,000 disputes, where 12,000+ of those fell within their scope. The report has some stats about their work, how many times they disagreed with the platform and overturned the result, and more.
DSA User Support Guide also by the Appeals Centre; good breakdown of your rights under the Digital Services Act. The platforms are supposed to tell you that orgs like the Appeal Centre exist, but somehow still don't, and many people don't know their rights. Hold them accountable! Know your rights and make use of the newly established bodies. Under Article 20 of the DSA, users must be able to lodge complaints, free-of-charge, against decisions taken by the platforms within the last six months.
papers, studies and similar publications
Dark Patterns in AI Chatbots - self explanatory; basically about design and interaction/output choices that maximize usage and data collection, lie about the capabilities and emotional intimacy etc. I learned a new term: Privacy Zuckering! Also made me read this about Gemini encouraging a guy to kill others, steal a mannequin, and then kill himself.
Arbeitspapier Identifizierbarkeit - German BayLfD summary and interpretation around identified and identifiable personal data in edge cases/gray areas, especially around pseudonymous data. What means to identify "count"? Not just your own! There's a difference between relative/subjective identification and absolute/objective identification. Sidenote: Love that they recommend RSS-Feeds or a Mastodon Account to keep up to date on legislation in this.
From intent to action: the leaders' guide to building AI-powered workplace - paper sponsored by Kyocera and done by Economist Impact, based on a survey of 639 senior executives conducted in October and November 2025, with in-depth interviews with businesses and "thought leaders" in AI, digital transformation and workforce strategy. So... take it with a grain of salt, it is very corporate and very incentivized to be pro-AI in the workplace. Their key findings show that they want more investment, more adoption. But: Despite the "propaganda" (so to say), it exposes a lot of weaknesses everyone is already talking about in the workplace. To name one thing I scoffed at: Page 12, the fact that so many measure ROI of AI use in vague "employee productivity", which is probably just increased output or increased closed cases, without looking at the quality. Sad. 4% are not even measuring any ROI for it!
Our Data After Us - paper by the CNIL about our digital remains. Covers questions like: Do you want the content to remain after your death? Who gets to have access and manage it? Should that person delete it, or should the platform automatically delete it? Should your remains be used to train an AI to impersonate you to help your loved ones? There seem to be age and gender differences to these answers.
You Trust Your Chatbot With Everything - Should You? - paper by Theodore Christakis from AI-Regulation.com. The findings are as expected: Every major provider now trains on consumer chats by default, providers typically reserve safety and abuse-prevention uses and feedback actions to override the training opt-out, and they all reserve the right for humans to read the conversation. The author suggests a "Sealed Mode" where the default settings/options constrain reuse and human access, allows no training, has no advertising, little personalization, and cryptographic hardening. In my view, it could be a good first step, but I fear in practice, it would be bastardized, as meaningless and misleading as Incognito Mode in browsers has been. Ideally, the things of a Sealed Mode should be the default you can then opt out of one by one, and it can be legislated so. We have seen that hidden settings within different menus and specific modes you have to first know about and then turn on do not help the average user, since they are never actively prompted about them or told about them by the company. This stuff only aids a risk-transfer from controller to data subject. So do not offer a silly little compromise - make them default, and do not allow it to cost anything. Choosing between payment or privacy sucks. We should sometimes ask ourselves: If LLMs are just another tool, would I want Microsoft to always have access to and review my Excel sheets? Of course not! So why should we accept this here?
At times, the author is too timid for me ("Yet the purpose of adopting this prism is not to export the GDPR as a universal template, nor to argue that the world should converge on European legal categories of individual control." hey, why not? We don't have the Brussels Effect for nothing; privacy legislation worldwide has been shaped by the GDPR, one example being Brazil!). Favorite chapter was the second one (Ghost in the Machine), as it goes in on how incomplete and lacking the warning labels are, together with how contradicting they are when everything else encourages you to freely share anything. Least favorite are the parts where chatbots are asked to answer something; I am sorry, but I will never see these as genuine, truthful, verifiable answers. This is treating them as a conscious employee that an regurgitate internal policies, not a probability machine who can be nudged to give specific output.Gewalteskalation als System: Nihilistic Violent Extremism in Deutschland - German paper on NVE that's mostly done by children and teens, who connect online over misanthropic and nihilistic tendencies and then see extreme violence and vandalism as the only way forward. Not always far-right or incels, but often. The paper explicitly mentions the Com network, 764, MKY and NLM. Aside from Telegram, Discord is the biggest place for it. I was surprised how lax and wide the definition of violent extremism is (imo, that would make a significant portion of the population violent extremists), and I think the way the authors narrow it down a bit is a good attempt.
cases
28.05.2026 – 26 O 869/26 aka the big one currently making the rounds about Google being responsible for the AI summary output. It will be interesting to see how that progresses and if it will be overruled.
This one for noyb.
books
- Don't know if it counts as it is a web format, but I finished reading 17776 by Jon Bois.
In total, that is roughly ~ 350 pages, if we count an online article as two pages on average; difficult to judge for 17776, I'd put it as 40 pages, maybe.
Reply via email
Published