Cyber Territories - Dispatch #2
Signal #2.1 A Dutch court has upheld the ruling that requires Meta to offer chronological feeds. This is a setback for algorithmic dominance, but a victory for user autonomy.
Source:
[Reuters - Dutch court upholds ruling forcing Meta to offer chronological feeds] (https://www.reuters.com/legal/litigation/dutch-court-upholds-ruling-forcing-meta-offer-chronological-feeds-2026-03-10/)
Dispatch:
An Amsterdam court has ruled that Meta must provide Dutch users with a chronological feed, an option not currently promoted on Facebook and Instagram. Meta's business model is based on the principle of algorithmic personalisation, meaning that the longer users stay on the platforms, the more adverts they see. Algorithms are designed to collect data on user behaviour on an ongoing basis.
A chronological feed disrupts this cycle by presenting content in the order in which it was published, rather than according to Meta's predetermined expectations. This restricts Meta's capacity to predict and influence user behaviour.
Algorithms have been found to have a tendency to reinforce echo chambers and polarising messages. A chronological feed disrupts this mechanism, offering users a more diverse range of opinions and information.
Reflection:
Which organisations will be responsible for shaping the information architecture of the future? In the ongoing battle for control of information, the key question is who will emerge victorious: will it be the platform owner, the machine, or the user?
§§
Signal #2.2 The news that an AI agent had hacked McKinsey's internal chatbot in just two hours is a wake-up call for cybersecurity, governance and trust in AI systems.
Source:
[The Register - AI agent hacked McKinsey chatbot for read-write access] (https://www.theregister.com/2026/03/09/mckinsey_ai_chatbot_hacked/)
[CodeWall - How We Hacked McKinsey's AI Platform] (https://codewall.ai/blog/how-we-hacked-mckinseys-ai-platform)
Dispatch:
An autonomous AI agent has breached McKinsey's internal AI platform, which is used by more than 40,000 employees for strategy, client research and document analysis. The agent was granted full read and write access to the production database. This exposed millions of strategic, M&A and client-related conversations.
The agent selected McKinsey itself as a target, based on their public responsible disclosure policy. The attacker could potentially have falsified or leaked financial models, strategic advice and confidential data without users or administrators noticing.
Reflection:
In light of the recent breach of security by an AI agent at McKinsey, a leading consultancy firm, how can we be sure that our AI systems are secure? In what manner are we disclosing information regarding our AI vulnerabilities? What is our preparedness and our ability to communicate effectively in such situations?
§§
Signal #2.3 The increasing uniformity of thought and writing is indicative of a silent erosion of cognitive diversity, which is being driven by AI.
Source:
[USC Dornsife - AI may be making us think and write more alike] (https://dornsife.usc.edu/news/stories/ai-may-be-making-us-think-and-write-more-alike/)
[StudyFinds - AI's Role in Our Lives Goes Beyond Writing. [It may be quietly reshaping how we think.] (https://studyfinds.com/ai-may-be-quietly-reshaping-how-we-think/)
Dispatch:
Researchers at the University of Southern California have expressed concerns that large language models (LLMs) are not only refining our texts, but also subtly standardising our thoughts, reasoning and opinions.
Research indicates that individuals who utilise AI for text composition or idea generation exhibit diminished individual style, reduced creative distinctiveness, and increased language and thought pattern uniformity. Furthermore, it appears that we unconsciously adopt the AI's framing and opinions without realising it.
While AI users generated a higher volume of ideas, these ideas exhibited a lower degree of originality and greater similarity to one another. Brain scans revealed a decrease in neural activity in regions associated with memory, concentration and creativity. AI has been shown to promote binary, logical reasoning, while inhibiting intuitive and contextual thinking. These are, however, the very skills required for innovation.
Reflection:
To what extent is our thinking influenced by AI in our daily interactions? What are the potential consequences of subjecting all our texts, ideas and opinions to the same algorithmic filters? Which of these factors constitute good thinking: the algorithm, the market, or human decision-making? In what ways might we design artificial intelligence that serves to enhance our creativity and cognitive diversity, rather than diminishing it?
§§
Signal #2.4 Meta and TikTok have been criticised for allowing harmful content to proliferate on their platforms. The algorithms used by these companies prioritise anger, polarisation and engagement over safety, which has led to concerns about the content that users, particularly younger ones, are exposed to.
Source:
[BBC News - Meta and TikTok allow harmful content to proliferate following evidence of outrage, say whistleblowers] (https://www.bbc.com/news/articles/cqj9kgxqjwjo)
Dispatch:
A BBC investigation involving more than a dozen whistleblowers and internal documents reveals that Meta and TikTok deliberately allowed and even amplified harmful content on their platforms.
This was after internal research showed that anger, polarisation and controversial topics significantly increased user engagement.
Internal Meta documents demonstrate that content which contravenes established moral standards, fosters animosity or incites violence is found to generate a significantly higher volume of bullying, hate speech and violence in comparison to other posts.
The algorithm interprets this as an indication that users are interested in seeing more of this type of content, and accordingly, it delivers it to them.
Whistleblowers have confirmed that Meta executives, under pressure from falling share prices and competition with TikTok, gave instructions to allow harmful content, including misogyny, conspiracy theories and violence, because this generates more advertising revenue.
An internal employee presented the BBC with dashboards that appeared to show TikTok prioritising complaints from politicians regarding cartoons over complaints concerning sexual blackmail, terrorism and child abuse.
Social media companies assert that they merely reflect society. However, these internal documents indicate that they are complicit in the normalisation of extremism, hate and disinformation.
Reflection:
How must we consider the implications of algorithms that are designed to amplify polarisation. Is this akin to the potential dangers associated with tobacco products, or worse? Why not request that these companies disclose the methodology behind their algorithmic content selection processes and the societal impacts thereof? To what extent are we ourselves aware of this manipulation? How do we identify the emotions and behaviours platforms evoke in us, and the reasons why?
§§
Signal #2.5 The Dutch Data Protection Authority issues a warning: It has been demonstrated that AI chatbots are unreliable and biased sources of voting advice. This could potentially compromise the integrity of the electoral process.
Source:
[Security.NL - AP reiterates advice not to use AI chatbots for voting advice] (https://www.security.nl/posting/928343/AP+adviseert+op+nieuwe+dat+AI-chatbots+niet+te+gebruiken+voor+stemadvies)
Dispatch:
The Dutch Data Protection Authority (AP) wishes to reiterate its advice against using AI chatbots for the purpose of voting advice. This advice follows research which revealed that such chatbots, including ChatGPT, Claude, Gemini, Grok and Mistral, provide unreliable, arbitrary and biased advice.
The research indicates that chatbots systematically disregarded local political parties, even when the user's input precisely aligned with the manifesto of such a party. The AP has stated that this undermines the integrity of free and fair elections and has called on both voters and developers to exercise caution.
The chatbots consistently recommended the same parties, regardless of the user's input, whilst other parties were systematically overlooked.
It is not clear why a particular recommendation is given to a user. It is important to note that the chatbots rely on unverifiable training data and do not provide clear source attribution.
If a significant proportion of undecided voters are influenced by such a chatbot, it could potentially result in an outcome that does not accurately reflect the actual state of democracy.
Reflection:
What measures can be implemented to ensure that AI-generated voting advice does not serve to reinforce existing political biases, but rather provides objective information to voters? In what ways can we form our own opinions in an era of algorithmic influence?
In what ways can journalists contribute to the fight against misinformation spread by AI chatbots?
What metrics can be used to assess the impact of AI on political preferences and democratic processes?
Synthetic image/AI-generated
This blog is written by Patrick Lacroix in a personal capacity. AI tools are used for research, structuring, drafting and language support. All content is selected, verified, and edited by the author, who retains full editorial responsibility.