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 external links with “➚” and all potentially paywalled links with “🔒”.
Well, that was quite a week for AI Ethics and science: OpenAI claimed to have solved a decade-old math problem with a new “internal model”. Coincidentally, it started working on this problem just as two researchers were finishing their work on the issue, raising some suspicions and many questions.
The issue is not who published a solution to the problem first, but why OpenAI can’t say if the researcher’s private drafts made it into the training material.
We looked at this case in “Is OpenAI Taking Everyone for Fools?”.
Other than a fight with academia, OpenAI has released their new GPT-6 model “Astra”, which caused CEO Sam Altman to announce the arrival of AGI.
Meanwhile, new details have emerged about OpenAI’s agents flooding a German website and transforming it into a messaging board for other agents (➚🔒The Verge). OpenAI says that “Claims that our Legal team discouraged investigation of the incident are false”.
OpenAI’s legal team is certainly busy, as the company faces 30 new complaints on behalf of victims and families connected to the Tumbler Ridge mass shooting. The new complaints claim that OpenAI’s Chief Global Affairs Officer told staff to stand down and not contact the authorities (➚Tech Crunch).
The involvement of Palantir in the British Health System has been controversial for a while, but now The Guardian reports that Palantir’s entanglement with the NHS may have caused tens of thousands of patients to withdraw their consent from using data in research projects (➚The Guardian). UK Health Innovation Minister James Frith has expressed worries that mistrust of the US tech company will affect “people’s willingness to share data with the NHS”.
Minnesota’s ban on AI-generated fake nude images will remain in effect after a federal judge declined to block the law at an early stage of the case brought by SpaceX AI while the case is being argued further.
SpaceX AI argues the measure restricts free speech protected by the U.S. constitution, while Minnesota claims that the law is narrowly tailored to curb the spread of nonconsensual sexual imagery created with AI (➚Reuters). SpaceX has now filed an interlocutory appeal in the case (➚Case 0:26-cv-03425).
While the US administration is pushing for more data centres, Republican’s in Texas and other states appear to change their mind on the issue: Republican candidates for the midterm elections in at least five states have adopted a tougher rhetoric on data centres, Reuters reports (➚Reuters).
Anthropic Might have settled the massive copyright lawsuit over their use of books as training material, but it looks like authors are not the ones benefitting from the settlement. While authors were promised $3000 per book as part of the settlement, they will need to split this money with the publishers, with some publishers aiming for a 50/50 split (➚🔒New York Times).
Six European countries have rolled back digital-first learning in schools to reintroduce physical paper books, notebooks, and handwriting, with Norway looking to restrict the use of AI in primary education (➚Teachers Updates).
A study by Yannic Meier published in the Journal of Science Communication looks at how social media users react to science content.
“The overarching goal of this study was to shed light on incidental exposure (IE) to science content on social media to better understand its process and to learn about any differences or commonalities to IE to political news using a user-centred approach.”
The study looks at questions of trust and the implications for science communication.
“The results of the present study imply that IE to science content can be associated with knowledge acquisition and becoming aware of new topics. However, the findings also indicate that the range of possible consequences of IE to science content is very broad. These consequences can range from experiencing no consequences at all to influencing opinions and voting decisions.”
Meier, Y. (2026). User perceptions of incidental exposure to science content on social media — a qualitative study. JCOM 25(6), A01.
If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.
