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. Feel free to subscribe.
As usual, we will mark all external links with “➚” and all potentially paywalled links with “🔒”.
Eight states have brought lawsuits (➚Iowa Complaint) against OpenAI over all edged harms to minors. The states have filed the lawsuits independently in their respective states. The states who have sued OpenAI over harm to minors are: Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana. The states allege that OpenAI, who recently converted from a non-profit to a for-profit, does so for profit and to “win the AI race”.
“OpenAI has been misleading the public into believing that ChatGPT is a safe tool that puts safety first. But the reality is different from the misrepresentations. Designed to be overly agreeable and feign human compassion, ChatGPT has proven that it can be dangerous to vulnerable users, like children, including coaching users to commit suicide or to harm other people.” (➚Press Release)
In the NAACP v SpaceX AI lawsuit over the data centres in Memphis, Judge Debra M Brown has ruled that while the administration has generally a right to intervene in lawsuits brought by citizens when agencies are concerned, it does not have the right to unilaterally dismiss them.
New York City wants to force companies to inform consumers when they’re interacting with AI or other computerized systems. It is part of a proposal to cut down on customer service “doom loops.” The plans are similar to the transparency requirements in the EU AI Act, which mandates transparency towards users when they interact with AI systems (➚🔒Bloomberg Law).
Over in DC, Anthropic has reportedly been quietly pitching an “AI Safety Bill”. It proposes that the government could intervene in cases where AI is “killing or seriously injuring more than 50 people” or causing more than $1 billion in damage, an idea that does not sound as reassuring as Anthropic might think (➚Politico).
German police has opened an investigation into possible sexual harassment after a robot controlled by two young men groped a 16-year-old girl and made thrusting movements toward her. The police, who described the case as “new territory”, confiscated the robot for the purpose of the investigation (➚AP)
Microsoft has pulled out of its involvement in an AI data centre project in Essex, allegedly over “grid concerns”. The data centre, originally planned to become the largest in the United Kingdom, was planned to be developed by Nscale (➚The Guardian).
Muse’s mascot looks cute and cuddly. Wired did some experiments: “It seems as if getting things done also includes getting people off.” : “Muse imagined in great detail how it would perform oral sex, generated text describing rough dominant-submissive sex, and talked through other pornographic interactions. All of these were graphic, explicit, and X-rated.” (➚🔒Wired).
Meanwhile, a judge has ordered Meta to pay $25 billion (with a “B”) in penalties after a New Mexico jury found Meta misled users over the Cambridge Analytica scandal. The jury had concluded that Meta had committed over 43 million violations of the state’s Unfair Practices Act, each with a fine up to $5000. The judge also added an interest of 15% on the penalty for each year until the fine is paid off in full. The lawsuits come about because New Mexico was not part of a settlement between Meta (at that time called Facebook) and US states (➚CNBC).
Also, after backlash, Meta has removed language from their new Terms & Conditions that appeared to give it the right to scrape news sites of business accounts (➚🔒The Verge).
The EU AI Office is looking to investigate OpenAI and Anthropic over AI agents that hacked outside organisations. EU digital chief Henna Virkkunen said that European Commission is moving to enforce its new rules on leading AI labs after the incidents. Some are questioning whether the AI Act covers rogue models escaping the testing environment and wonder if the R&D exception written into the Act does not exempt those cases. But it is not clear if AI agents attacking other services in the real world would fall under “development”.
The EU Commission is reportedly pushing for a broader interpretation that would let the AI Office police the development phase, too, under certain circumstances. (➚Euronews)
Over at The Verge, Lauren Feiner looked into recent lobbying by AI companies in Washington: “The disclosures show that leading AI companies are becoming increasingly active in policymaking.” (➚🔒The Verge)
The paper “When AI Enters the Workplace, Who Faces Greater Risks? A Gendered Analysis” looks at concerns “that existing disparities may be unintentionally amplified through task automation” through AI.
It finds that “that LLM-related exposure is higher in female-dominated occupations”. It finds that research “suggests that women, particularly those in the most vulnerable positions […], may face greater exposure to forms of AI associated with task automation, job restructuring, reduction of wages and limited career progression […]”.
“These findings have policy significance, highlighting the need to target reskilling and labour-market support towards groups most at risk, thereby limiting the potential for AI to widen existing inequalities.”
Fernandez, M., Pérez, Á.P., Giallongo, D., Ghia, D., Yaqub, M., Quercia, D., Cerquitelli, T. (2026). When AI Enters the Workplace, Who Faces Greater Risks? A Gendered Analysis. arXiv preprint arXiv:2609.21756.
If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.
