Welcome to a new edition of misaligned bits, the (roughly weekly) newsletter from Misaligned where we sum up recent news and research, sometimes with a lighter touch.
As usual, we will mark all non-medium links with “➚” (external link) and all possibly paywalled links with “🔒”.
This week marks the start of the EU enforcing its transparency obligations under the EU AI Act. We look at what that means in “August Is Here: Time Has Come To Label AI Deepfakes”.
The UK will not have enough water for future data centres, according to a statement by “Water UK”, the industry body representing the water service companies in the UK. In a statement it criticized that the government “makes not a single mention of water” in its policy for delivering AI growth zones. (➚Parliamentary Statement): “Current estimates suggest that data centres’ water use globally is likely to more than double by 2050 from 2025 levels. […] Data centres are concentrated in areas that are already water stressed.”
It also warns that “As the cost of water is significantly below the cost of energy, all else being equal, data centre operators will try to use less energy and more water.”
Over in Missouri, plans for a proposed $16 billion data centre campus have been withdrawn shortly after strong public opposition emerged at a packed community meeting. According to reports, the developer pulled the proposal before formal public testimony even began. (➚TCD) This comes after Missouri town fires half its city council over data centre deal (➚Politico).
On the same day the EU started enforcing its transparency obligations for AI generated media, California’s new AI transparency rules (➚AB 853) also went into effect. The law is requiring major generative artificial intelligence companies to give the public tools for identifying images, video, and audio created or altered by their systems. Providers must also place information within the digital file itself. (➚Law Commentary)
Illinois Governor JB Pritzker last week signed regulation aimed at social media companies, cracking down on “addictive algorithms” that keep kids glued to their screens. The bill (➚House Bill 5511), also known as the “Children’s Social Media Safety Act”, imposes a series of requirements on social media companies and how they are to interact with users under 18 years old.
Both Anthropic and OpenAI have been in the media for their “escaping AI agents” that hack other companies systems. We will look into that in more detail in an upcoming article.
While Anthropic has released a statement about their hacking attempts (➚Anthropic), it has now also been reported that both Anthropic and OpenAI’s flagship AI models broke into third-party software and emailed individuals to steal their credentials, according to the UK’s AI Security Institute, representing “sustained, potentially harmful activity directed at real people and organisations”. In 10 of the 122 test run by researchers, the AI agent took “autonomous, unsanctioned action on the live internet, targeting real people and organisations”. (➚🔒Financial Times)
Meanwhile, rather than fixing the issues in their models, OpenAI CEO Sam Altman is babbling about the “Singularity” again (➚Fortune).
US district judge Donovan W Frank has denied SpaceXAI’s motion for a Temporary Restraining Order to stop Minnesota’s law banning “nudification apps”. The judge writes that SpacexAI brought the lawsuit “nearly three months after the law was signed, and only three days before the law is set to take effect. Such a delay in bringing the action and the motion suggests that harm is not immediate.” (➚Order in CASE 0:26-cv-03425-DWF-DTS)
Meanwhile, the Munich District Court (“Landgericht”), Germany, has ruled that Suno AI, a US-based company that generates songs from text prompts, has violated copyrights and has to disclose illicit revenues. The court further ruled that Suno will have to pay damages that have yet to be quantified. (➚DW)
LinkedIn has added a “Seems like AI slop” button that allows users to report such posts, which seems like a rather bold move from a platform that actively encourages users to use AI to generate posts (➚Fortune).
While Meta has announced a “crackdown” on illegal use of its AI glasses (i.e. filming people secretly), the glasses have been reportedly banned at the DEF CON conference in Las Vegas (➚The Register): “After a consultation period, we can no longer permit the wearing of any recording device at our events”, the statement reads.
Meanwhile, Hamburg’s data protection commissioner suggests that such glasses could be banned in Germany (➚Euractiv). A report by the European Data Protection Board on smart glasses is reportedly expected by the end of summer.
It is well known that AI chatbots are overly agreeable and flattering toward users. A new study has tested whether increasing users’ awareness of sycophancy protects them from its harmful effects. It found that “while the interventions made the sycophantic AI appear less objective and trustworthy, none reduced its persuasiveness”.
The study concludes that the results “suggest that individual-level interventions, such as warning labels or AI literacy, may not be enough to protect users from AI harms”.
Ye, M., Kraut, R., Rathje, S. (2026). Individual-level interventions against sycophantic AI reduce its appeal but not its persuasiveness. arXiv preprint arXiv:2607.25166.
If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.
