Bias wherever you look, August comes closer, Palantir loses contract in France, police reportedly used AI to manufacture evidence and ChatGPT can be easily tricked.

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 “🔒”.

Regulatory bits

The EU Parliament has approved the final text (➚European Parliament) of the updated AI Act, mainly consisting of changes in the implementation deadlines. Before the law can enter into force, it still needs to be adopted formally by the Council. Nevertheless, more provisions of the AI Act will start to apply on 2 August 2026, including the transparency rules that will obligate providers to disclose interaction with AI based systems and labelling rules for deepfakes.

The EU commission has recently published its guidelines for those transparency rules. (➚European Commission)

The governor of the US state of Vermont signed bill H.816 last week, effectively banning the use of AI chatbots for mental health services. The bill states “It is the purpose of this act to safeguard individuals seeking mental health services in Vermont from psychological harm, including death by suicide, by ensuring that these services are delivered by mental health professionals and not independently by artificial intelligence systems.” (➚Bill H.816)

Palantir bits

The French domestic intelligence agency has cut its contract with Palantir and replaced it with local company ChapsVision as part of a drive by European countries to reduce their dependence on US tech firms. (➚🔒Financial Times). This is the second significant contract Palantir loses in Europe after the mayor of London vetoed a £50m contract with the Metropolitan Police, a decision Palantir plans to challenge in court. (➚The Guardian).

British bits

A Derbyshire police officer is under criminal investigation over the alleged use of AI to “create evidential material in a number of cases”. Derbyshire police told the Financial Times that “a criminal investigation has been launched into an allegation of perverting the course of justice after the alleged use of AI.” (➚The Guardian)

Trickery

According to researchers from Mindgard, the latest public version of ChatGPT can be easily tricked into generating sexualised images or depict scenes of graphic violence with a simple prompt. OpenAI has told the BBC they have introduced (or will introduce?) additional safeguards against this type of prompt (➚Mindgard).

US political bits

Dean Ball, a former top White House official of the Trump administration and main author of Trump’s “AI Action Plan” is going to join OpenAI as “as head of strategic futures”, whatever that is. (➚Politico)

Meanwhile, the US Department of Justice, intervening in the case brought by NAACP against the SpaceX data centre in Memphis, is trying to get the case dismissed citing “national security concerns”. It also argues it has the authority to stop environmental lawsuits brought by citizens. It appears that running dozens of gas turbines without permit relates to “national security”, likely a first. (➚🔒New York Times). As a side note: This is the second time the US government has intervened in a lawsuit involving xAI/SpaceX this year. They also intervened in SpaceX’s lawsuit against the Colorado AI regulation that wanted to ban discriminatory algorithms.

Science Bits

Implicit bias against people with intellectual disabilities

A paper published in the “Disability and Health Journal” investigates the presence of implicit bias in Large Language Model (LLM)-based chat AI models directed toward people with intellectual disabilities (ID).

(Note: The full study is behind a paywall)

“These implicit biases are considered within the context of past discrimination towards people with ID and highlight the need for diligence against implicit bias towards people with ID in AI development. This research underscores the importance of assessing and mitigating implicit bias in decision-making technologies to prevent future societal harm.”

Karly V. Coffey, Gloria L. Krahn, John P. Hanley, Jacob E. Neely (2026).
​Identifying implicit bias in LLM-based chat AI toward people with intellectual disabilities. Disability and Health Journal. ISSN 1936–6574

Bias in historical narratives

In a paper published in “PNAS Nexus”, Matthew Shu et al. look into AI chatbot’s “capacity to influence opinions through subtle, latent ideological framing”.

While most research has focused on false information given by LLMs, the researchers come to the conclusion that their “findings demonstrate that the use of AI for learning history can influence opinions through both intrinsic and intentional framing mechanisms, even when the content remains factually accurate”.

“[…] Our findings identify the persuasiveness of AI’s latent biases and the implications of using AI to obtain factual information. The presence of latent biases means that simply using chatbots to learn about the past can influence people’s opinions at a magnitude comparable to prompting-biased AI material”

Matthew Shu, Daniel Karell, Keitaro Okura, Thomas R Davidson (2026). How latent and prompting biases in AI-generated historical narratives influence opinions, PNAS Nexus, Volume 5, Issue 3

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.