Manifest fallout, legal bits, AI Psychosis and relying on AI assistants could be bad for your performance.

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 “🔒”.

Manifest bits

More than 200,000 people have signed two petitions urging the UK government to sever its ties with the company Palantir after it released its “mini Manifesto” (➚The Guardian). The company hold contracts with the NHS and the military, and has been recently in talks with the Metropolitan police.

Meanwhile, the UK Science, Innovation and Technology Minister Kanishka Narayan has cast doubt on OpenAI’s stated reason for pausing a major infrastructure project, instead suggesting the company’s finances could be behind the decision. (➚Politico)

Legal hallucinations, again

The law firm Sullivan & Cromwell told a US federal bankruptcy court that a major filing it made in a high-profile case contained multiple “AI hallucinations”. (➚🔒Financial Times) According to S&C, at least five high-level partners have been assigned to this bankruptcy case, who all appear to have not recognized the error.

Regulatory bits

OpenAI is backing a bill that would limit liability for AI-enabled mass deaths or financial disasters. The company testified in favour of an Illinois bill that would limit when AI labs can be held liable, even in cases where their products cause “critical harm.” (➚🔒Wired)

Legal bits

Another interesting lawsuit is the one filed by Nippon v OpenAI that has been filed last month: The plaintiff in this case is a company that was sued by an individual for various claims, and that individual allegedly used ChatGPT to create legal filings.

Plaintiffs ask the court to hold that “that OpenAI has violated state law by practicing law in the state without a licence”. They also ask the court to “permanently enjoining OpenAI, from engaging in the practice of law in the state of Illinois”. While in itself a rather small claim in itself, it raises the broader question of what does and what does not “legal advice” by a chatbot. (➚Case: 1:26-cv-02448)

According to the WSJ, the US justice department has told French authorities it will not facilitate their efforts to investigate Elon Musk’s X. The DoJ says that the “investigation seeks to use the criminal legal system in France to regulate a public square for the free expression of ideas”. French authorities investigate X for dissemination of CSAM and Holocaust denial (a criminal offence under French law). (➚🔒WSJ)

Meanwhile, Elon Musk did not attend a voluntary interview he was summoned to appear at in Paris today, according to French authorities.

As we mentioned in our article on the Grok scandal, Apple’s app store guidelines do not apply equally to all. However, a federal judge has now ruled that it was the Trump administration who illegally coerced Apple into removing an ICE-tracking app. The government, the judge writes, “demanded, rather than requested” that the platform censor the plaintiffs’ speech. (➚Courtlistener)

Science bits

In their study ”AI Psychosis” in Context Luke Nichols et al. note that extended interaction with large language models (LLMs) has been linked to the reinforcement of delusional beliefs, yet most empirical work evaluates model safety in brief interactions.

They come to the conclusion that “the results suggest that delusional reinforcement by LLMs reflects a preventable alignment failure” and “these findings indicate that accumulated context functions as a stress test of safety architecture, revealing whether a model treats prior dialogue as a worldview to inherit or as evidence to evaluate.”

Nicholls, L., Hutto, R., Soto, Z., Morrin, H., Pollak, T., Korpan, R., Carmichael, C. (2026). “AI Psychosis” in Context: How Conversation History Shapes LLM Responses to Delusional Beliefs. arXiv preprint arXiv:2604.13860.

Grace Liu et al. looked at the impact of AI assistance on independent performance. They note that “current AI systems are fundamentally short-sighted collaborators” which does not track progress or “prioritizes the other person’s growth over immediate results”. They identified two key consequences of AI assistance: reduced persistence and impairment of unassisted performance.

“Across a variety of tasks, including mathematical reasoning and reading comprehension, we find that although AI assistance improves performance in the short-term, people perform significantly worse without AI and are more likely to give up.”

Liu, G., Christian, B., Dumbalska, T., Bakker, M.A., Dubey, R. (2026). AI Assistance Reduces Persistence and Hurts Independent Performance. arXiv preprint arXiv:2604.04721.

Follow and subscribe

If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.

You can now also subscribe to this newsletter directly with your email.