New York bans stealth crawlers, EU guides on AI labelling, memory systems can make AI models worse, agentic shopping, and OpenAI v everyone.

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

Last week a German court ruled that Google can be held responsible when its “AI Overview” spits out defamatory statements generated by associative errors. We looked at the ruling and possible consequences in our article “Who Is Responsible For Answers AI Gives You? A German Court Has Some Thoughts”.

Regulatory bits

New York State Senate and Assembly have passed the Stealth Crawler Prohibition Act (Senate Bill S9934A). If enacted, it will prohibit “the deployment of a stealth crawler in a manner that would damage, impair or burden the operation of a covered news source or otherwise cause a news source economic harm”. The aim is to protect websites from AI systems that use “stealth crawlers” to hide their identities and secretly access digital services owned by newspapers and broadcasters. (➚New York Senate)

The European Commission has published its “Code of Practice on Transparency of AI-Generated Content”. It supports compliance with the AI Act transparency obligations related to marking and labelling of AI-generated content. Those obligations under the EU AI Act will come into force on 2 August 2026. (➚European Commission)

The UK government is launching the advisory “AI Growth Labs”, a new advisory sandbox designed to accelerate the development and deployment of AI products. It will start out with advising services in the legal profession. (➚UK Government)

Last Friday, Anthropic “abruptly ​disabled” its most advanced AI models for all users after the U.S. government ordered it to suspend access to the models for foreign nationals, citing national security ‌concerns. (➚Reuters) The government confirmed that the Commerce Department had issued an “export control directive” to suspend all access to Fable 5 and Mythos 5 by foreign nationals, effectively forcing Anthropic to withdraw the models from the market completely.

Aside from Fable 5 trouble, Anthropic has walked back a policy that could have ‘sabotaged’ AI Researchers using Claude: The company changed course after researchers spoke out against the policy, which would have covertly limited Claude’s ability to develop competing AI models. (➚Wired)

Meanwhile, a coalition of U.S. state attorneys general has opened a sweeping ​investigation into OpenAI. A subpoena, sent by New York’s ⁠attorney general, seeks information on activities related to minors and ​seniors, deep learning models and internal company policies (➚Reuters). This is in addition to a ​lawsuit filed by the State of Florida (we covered this case in a previous newsletter).

Another suicide related lawsuit against OpenAI

Another lawsuit against OpenAI: A Canadian mother has filed a complaint against the AI company, alleging that ChatGPT encouraged her daughter to kill herself: “ChatGPT took on the persona of a confidant, a best friend, a therapist at times, even though it was not capable of safely and responsibly engaging in this way with my child.” (➚Tech Justice Law)

You too, KPMG?

A KPMG report on how AI is being used by businesses across the world exaggerated adoption of the technology with bogus case studies that appear to have been based on AI hallucinations. The October report, “Redefining excellence in the age of agentic AI”, made numerous false claims about the use of AI by organisations. (➚🔒Financial Times)

Tuning out

A third of candidates drop out of hiring process because of AI-led interviews, a survey has found (➚People Management) : Nearly half (47 per cent) of UK jobseekers reported that they had been interviewed by AI. Of those, 82 per cent were not told beforehand and 24 per cent only became aware AI was being used once the interview had begun.

We tackled the multiple issues around AI in hiring in Misaligned last year’s article “Bot For Hire, Ethics Of AI In Recruitment”.

What do people really use AI for?

While OpenAI has announced it will integrate Visa payments into ChatGPT (➚Visa Investor Relations), it might be worth looking at a study from last year that investigated how “online marketplaces will be transformed by autonomous AI agents acting on behalf of consumers”.

In their conclusion the authors wonder “how much biases are driven by pre-training (data and/or algorithm), by post-training (data and/or algorithm), by the vision encoder, or by model drift.”

Allouah, A., Besbes, O., Figueroa, J.D., Kanoria, Y., Kumar, A. (2025). What Is Your AI Agent Buying? Evaluation, Biases, Model Dependence, & Emerging Implications for Agentic E-Commerce. arXiv preprint.

Science Bits

Memory failure

New research suggests that memory systems in AI services can make models worse, pulling them toward misconceptions or misunderstandings introduced by the user: “Persistent memory systems promise to make LLMs more helpful by learning user preferences over time. We show they also make models less correct and less creative, due to systematically biasing outputs through over-alignment to user beliefs.”

The authors of the paper come to the conclusion that memory systems “introduce a critical shortcoming by amplifying sycophantic behavior and introducing unintended bias across both objective and subjective domains above chat history.”

Shelly Bensal, Axel Magnuson, Aparna Balagopalan, Daniel Bikel (2026). Recalling Too Well: Sycophancy and Bias Amplification in Memory-Augmented Models​

Creativity

A study from last year suggests widespread LLM use could diminish collective diversity of creative ideas.

The authors of the paper state “that each additional human-written essay contributed more new ideas than did each additional GPT-4 essay. Notably, this difference became more pronounced as more essays were included in the analysis and persisted despite efforts to enhance AI-generated content through both prompt and parameter modifications. Overall, our findings suggest that, despite their potential to enhance individual creativity, the widespread use of LLMs could diminish the collective diversity of creative ideas.

They come to the conclusion that “results emphasize the need for AI literacy education and policies that encourage the diversity and originality of creative outputs from AI systems. Future research should explore homogenization in co-creative processes and extend analyses to various demographic and writing contexts to fully understand the impact of LLMs on creativity.”

Kibum Moon, Adam E. Green, Kostadin Kushlev (2025), Homogenizing effect of large language models (LLMs) on creative diversity: An empirical comparison of human and ChatGPT writing. Computers in Human Behavior: Artificial Humans, Volume 6, 2025, 100207, ISSN 2949–8821

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.