AI is swallowing poison, AGI scares are back on the marketing menu, copyright questions again and LLMs are bad for your health.

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 “🔒”.

Misaligned Recap

As a quick reminder, can now also subscribe to this newsletter directly with your email.

Weird bits

OpenAI has acquired the TBPN podcast (➚OpenAI), which OpenAI describes the podcast as “one of the places where the conversation about AI and builders is actually happening day to day.”

The conflict of interest in running a podcast owned by a major player in the AI industry should be obvious to everyone, and can be seen as a marketing play from OpenAI to counter the current backlash against large AI companies.

Anthropic has “leaked” that it allegedly has a model in its lab that is “too powerful to be released” (NBC). This appears to be part of a new attempt to market AI as secretly more powerful than it is, even more so as the information was coincidently released at the same time Nvidia CEO Jensen Huang and Silicon Valley investor Mark Andreessen claim that “AGI” is already here. (Forbes)

Bit of copyright, again

Penguin Random House has filed a lawsuit against OpenAI after ChatGPT generated text and images from a German children’s book series that were “virtually indistinguishable” from the originals. (➚The Guardian)

Anthropic meanwhile is still racing to contain the fallout after accidentally exposing the underlying instructions it uses to direct Claude Code By Wednesday morning, Anthropic representatives had used a copyright takedown request to force the removal of more than 8,000 copies and adaptations of the raw Claude Code instructions. (➚🔒WSJ) (➚Anthropic)

While this is happening, the CEO of Anthropic says he “is not trying to convince Australia to change its mind on protecting the copyright of musicians, writers and other artists”. (➚ABC)

Meanwhile, the European Parliament held a press conference last months on the Parliament’s proposals to protect EU creative production in the age of artificial intelligence. (➚European Parliament)

Thousands of authors including Kazuo Ishiguro, Philippa Gregory and Richard Osman have published an “empty” book to protest against AI firms using their work without permission under the title “Don’t steal this book”. Copies of the work are being distributed to attenders at the London book fair this week (➚The Guardian).

For further reading on the current state of AI and copyright in various jurisdictions, we recommend the overview published by Lawdit in December 2025 (➚Lawdit).

Bad for your Health

A team from the University of Gothenburg, Sweden, performed a test on how quickly medical misinformation ca spread with the help of Chatbots: The team published a (non peer reviewed) paper on a faked disease, then waited for AI crawlers to read it.

Within weeks of uploading the information about the condition, attributed to a fictional author, major artificial-intelligence systems began repeating the invented condition as if it were real. Because of that, some other papers even started to cite it. (➚Nature) The test highlights how easily AI crawlers are tricked into spreading disinformation deliberately set up for them.

The largest user study of large language models for assisting in medical decisions has found that they present “risks to people seeking medical advice due to their tendency to provide inaccurate and inconsistent information”. (➚Oxford University) The study also revealed a two-way communication breakdown: Participants often didn’t know what information the LLMs needed to offer accurate advice. It also found that current evaluation methods for LLMs do not reflect the complexity of interacting with human users, and existing tests fall short.

The San Francisco-based startup “Legion Health” is planning to launch a one-year trial for Utah residents in which chatbots can serve as their psychiatrists and prescribe medications. Medical experts are not thrilled about the idea. Brent Kious, a University of Utah School of Medicine professor and psychiatrist, is warning that overreliance on artificial intelligence or automation could result in an “epidemic of over-treatment.” (➚🔒The Verge)

Private bits

A new paper published by the Bank for International Settlements looked into the biggest barrier to AI adoption in finance and unsurprisingly found it is the difficulty of AI to use datasets due to privacy or other regulatory reasons: “Key concerns include data privacy, quality and security, which are further intensified by third-party dependencies and market concentration among major service providers.”

The paper calls on financial authorities to provide clearer supervisory expectations on AI-related data usage. (➚BIS) There is the obvious danger that privacy regulation will be further softened to support AI adoption in the area.

Science Bits

A paper by Bean et al published in Nature describes the results of a randomized study where participants were asked to make decisions about a medical scenario as though they had encountered it at home.

The study found that “LLMs generated several types of misleading and incorrect information” and concludes that “that none of the tested language models were ready for deployment in direct patient care”.

Bean, A.M., Payne, R.E., Parsons, G. et al.Reliability of LLMs as medical assistants for the general public: a randomized preregistered study. Nat Med32, 609–615 (2026).

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.