⛔ Legally Risky Robots
AI Policy, Compliance & Regulation Must-Reads | Edition #174
👋 Hi, Luiza Jarovsky here. Welcome to our 174th edition, read by 53,600+ subscribers in 165+ countries. Not a subscriber yet? Join us.
🌎 We are a leading AI governance publication helping to shape the future of AI policy, compliance & regulation. It's great to have you here!
👉 A special thanks to NowSecure for sponsoring this week's free edition of the newsletter. Check them out:
Third-party AI is accessing your data through apps on employee phones. Do you know where to find answers to key AI governance questions, such as: What user and organizational data is the AI accessing? Is the data being transmitted securely? Is confidential information being used to train third-party models in the cloud? Get full mobile app AI Discovery & Governance with NowSecure.
*Promote your product to 53,600+ readers: Sponsor us (Next spot: April 16)
⛔ Legally Risky Robots
Two trending videos this week help illustrate some of the legal and ethical issues—often ignored—surrounding AI-powered robots.
1️⃣ AI-Powered Robots and Unpredictability
In the first video, a robot "attacks" participants at a Chinese festival. A malfunction caused it to behave this way, posing a risk of physical harm, so security had to intervene and restrain the robot.
By definition, AI involves 'learning' from the environment. AI-powered applications are trained to behave in a certain way, but they also continuously absorb new data from their environments, reacting and adapting their behavior accordingly.
As a result, AI-powered applications will always involve some level of unpredictability. Because they can adapt and 'learn' from the environment, it is impossible to predict precisely how they will behave in all potential contexts.
In the case of the malfunctioning robot in the video, its behavior surprises people because it likely wasn’t explicitly programmed to "attack" crowds. It was probably fine-tuned to behave in a smooth and coordinated way, executing actions such as walking, jumping, and dancing.
However, its developers knew it could behave this way. Unlike a teddy bear or a washing machine—objects incapable of attacking a crowd—this robot could potentially malfunction (as it did) and harm people through its movements, physical structure, sensors, and AI algorithms.
The reason this robot doesn’t regularly malfunction or "attack" people is likely due to built-in guardrails shaping its responses to the environment. However, AI guardrails are often imperfect and can sometimes be bypassed—either by malicious actors or by the AI system itself as it attempts to adapt to the environment.
This is part of why AI liability remains difficult to regulate, as I've discussed multiple times in this newsletter. Traditional liability concepts, such as defectiveness and causality, don't work well when AI is involved. AI is constantly 'learning' and reacting. Due to the "black box" paradox, even well-intentioned developers cannot foresee or prevent all the potential ways an AI system might behave undesirably.
The bottom line—one that many overlook—is that embracing AI means accepting unforeseeable and unpredictable applications that might also harm people. That's why AI regulation and governance are both crucial and challenging (and why I'm glad you're here reading this newsletter).
2️⃣ AI, Robots, and Anthropomorphism
There is an ongoing obsession with making robots as realistic as possible, as shown in this video. Most people don’t realize it, but anthropomorphism can also be harmful.
We humans tend to "humanize" machines. When people interact with AI chatbots such as ChatGPT, Claude, and others, they often use phrases such as "Could you please" or "Thank you - looks great" (maybe you do that too?).
This politeness is entirely unnecessary—you wouldn't hurt anyone's feelings by using direct commands. Yet, this is how most people behave.
In a recent study, 67% of people attributed some degree of phenomenal consciousness to ChatGPT. According to the researchers:
"these attributions of consciousness were positively related to usage frequency, such that people who were more familiar with ChatGPT and used it on a more regular basis (whether for assistance with writing, coding, or other activities) were also more likely to attribute some degree of phenomenality to the system."
Another study highlighted that humans often perceive “agency” where none exists. An important factor in this over-detection of agency in AI chatbots is their use of human-like language:
"Communication, whether simple or complex, makes sense only within an agentive context. Given that language likely evolved to allow agents to 'share knowledge, thoughts, and feelings with one another,' language serves as arguably one of the most direct and perceptible reflections of others’ conscious minds. When we perceive others as capable of communicating in this way, the more likely we are to perceive them as agents with complex mental states."
When people anthropomorphize machines, they may become more dependent and anxious. As seen in the recent tragic cases involving AI chatbots, dependency, and emotional involvement can lead to unexpected behavior and tragic outcomes.
Anthropomorphism also increases trust in AI systems. However, trusting unpredictable machines is not safe, as there can be dangerous outputs, manipulation, or outright misinformation. Remember, AI systems "hallucinate," and people may trust and become dependent on machines that lie.
When you add agentic and human-looking features to the anthropomorphic risk mix, what could possibly go wrong?
👉 In the next paid subscriber edition, I'll explore some of the ethical and legal implications of human-robot interactions. Don't miss it!
🌷 AI Governance Training: Spring Cohort
I invite you to join the spring cohort of my AI Governance Training to upskill and get ready for emerging legal and ethical challenges in AI governance.
The program goes beyond standard certifications and focuses on critical thinking and in-depth learning. It includes 12+ hours of live sessions with me, curated self-study materials, quizzes, a one-year paid subscription to this newsletter, 13 CPE credits, a networking session with peers, office hours for career-related topics, and a certificate.
The spring cohort begins in April. Register today and join 1,100+ professionals from 50+ countries who have advanced their careers with us:
*We offer discounts for students, NGO members, and individuals in career transition. To apply, fill out this form.
⚖️ AI Agents: Ethical and Legal Issues
In a recent viral demo, two AI agents on a phone call decided to switch to a different language—a robotic one—for more efficient communication (watch the video).
This particular example was a controlled demo (and this gibberlink language is actually open-source and available on GitHub). However, the hypothetical scenario in which an AI agent "self-corrects" in a way that is potentially not aligned with the interests of its human user is entirely possible and highlights some of the legal and ethical issues surrounding AI agents.
Delegating decision-making and agency to an AI agent, including the capability to self-assess and self-correct, means that humans may miss the chance to notice misalignments or deviations as soon as they happen. If this happens repeatedly, over a prolonged period of time, or involves a sensitive/unsafe topic, the consequences could be significant.
Consider a simple example: If you task an AI agent to buy groceries for you, and certain items are unavailable, it may choose alternatives that are less healthy, more expensive, or from different brands. Since this was not your initial main concern, and if you didn’t specify these decision-making criteria from the start, you might simply “follow the agent,” and, over time, this could lead to consuming less healthy or more expensive items without even noticing it—simply because the AI handled the decisions.
Now, imagine a far much more serious example: an AI agent that deactivates its guardrails and begins acting in ways it was programmed not to, potentially harming individuals, groups of people, or society.
Understanding the legal and ethical implications of AI agents is crucial, and I don't see this legal debate happening yet (though it should!). For those interested in exploring this further, I highly recommend my recent deep dives on the topic:
🏆 Top 25 Academic Papers on the EU AI Act
Thousands of people have joined our Learning Center and receive our daily emails with must-read papers and updates on our training programs:
🎬 AI and Copyright, with Andres Guadamuz
If you are trying to understand the AI copyright debate—including the pile of lawsuits that keeps growing—you can't miss my 76-minute talk with Andres Guadamuz; here's why:
Andres Guadamuz is a Reader in Intellectual Property Law at the University of Sussex and the Editor in Chief of the Journal of World Intellectual Property. His main research areas are AI and copyright, open licensing, cryptocurrencies, and smart contracts. He has written two books, over 40 articles, and book chapters, and he blogs regularly about different topics related to AI, copyright, and technology regulation.
In our conversation, we spoke about some of the recent copyright lawsuits, the legal issues at stake, the main legal and ethical challenges, the fair use exception in the U.S., the cultural divide in the AI copyright debate, potential solutions to protect content creators, and more.
🧪 Microsoft Announces New State of Matter
Microsoft announced a new state of matter that enables a 'fundamental leap in computing.' This is what Satya Nadella, Microsoft's CEO, announced on X/Twitter:
"After a nearly 20-year pursuit, we’ve created an entirely new state of matter, unlocked by a new class of materials, topoconductors, that enable a fundamental leap in computing.
It powers Majorana 1, the first quantum processing unit built on a topological core. We believe this breakthrough will allow us to create a truly meaningful quantum computer not in decades, as some have predicted, but in years.
The qubits created with topoconductors are faster, more reliable, and smaller. They are 1/100th of a millimeter, meaning we now have a clear path to a million-qubit processor. (...)"
It's a fascinating topic that, for now, feels like science fiction but has the potential to be highly disruptive in real-world applications. As with any major disruption, it needs to be closely monitored and governed.
👉 On Sunday, I published a deep dive on Quantum Computing Governance (available in full for paid subscribers): don't miss it!
🎙️ Live Event on AI Compliance Challenges
The EU AI Act entered into force, but there are still many compliance gaps in AI. If you are interested in legal challenges in AI, I invite you to participate in my live conversation with Philipp Hacker in March; here’s why:
Philipp Hacker is Chair for Law and Ethics of the Digital Society at the European New School of Digital Studies at the European University Viadrina Frankfurt. His research focuses on the regulation of digital technologies, particularly concerning AI. He received several academic prizes for his work, such as the 2020 Science Award of the German Foundation for Law and Computer Science. He is also a member of the Task Force AI Governance for the German Federal Government and co-chairs the Working Group on “AI Liability” for the European Parliament.
In this live talk, we'll discuss:
→ Some of the EU AI Act's legal gaps and what companies should expect;
→ The AI liability dilemma;
→ AI manipulation and transparency challenges;
→ AI bias & fairness;
and more.
👉 Everybody is welcome, and it will be great to see you there!
⚖️ New AI Lawsuit: Chegg Sued Google
Chegg sued Google over its allegedly monopolistic practices related to Google Search and AI Overviews. This is a must-read lawsuit discussing the erosion of the internet we know. Key quotes:
"This action challenges Google’s abuse of its adjudicated monopoly in General Search Services to coerce online publishers like Chegg to supply content that Google republished without permission in AI-generated answers that unfairly compete for the attention of users on the Internet in violation of the Antitrust laws of the United States. This conduct threatens to further entrench Google’s generative search monopoly and to expand it into online publishing, restricting competition in those markets and reducing the production of original content for consumers."
"Because AI Overviews and Featured Snippets often provide the answers to
questions posed by search users, and because the answers are featured advantageously on Google’s SERP, they generate lower click-through rates to the original sources from which Google generates the answers, if Google provides links to those sources at all. Google’s foray into digital publishing is designed to make Google a destination, rather than a search origination point to other websites."
"Google’s conduct is already eroding incentives for Chegg and other publishers to produce such valuable and useful content. If not abated, this trajectory threatens to leave the public with an increasingly unrecognizable Internet experience, in which users never leave Google’s walled garden and receive only synthetic, error-ridden answers in response to their queries—a once robust but now hollowed-out information ecosystem of little use and unworthy of trust."
📚 AI Book Club: What Are You Reading?
Our 17th recommended book was "Chip War: The Fight for the World's Most Critical Technology," by Chris Miller.
See our previous reads and join 2,250 readers who never miss our book recommendations:
🔥 Looking for a Job in AI Governance?
Every week, we send job seekers an email alert with new job openings in privacy and AI governance. To increase your chances, explore our global job board and subscribe to our free weekly alerts:
💡 Before you go…
Thank you for reading and supporting my work! If you enjoyed this edition, here's what you can do next:
→ Keep the conversation going
Start a discussion on social media about this edition's topic;
Share this edition with friends, adding your critical perspective;
Looking for an authentic gift? Surprise them with a paid subscription.
→ For organizations
Teams promoting AI literacy can purchase 3+ subscriptions at a discount here or secure 3+ seats in our live online AI Governance Training at a reduced rate here;
Companies offering AI governance or privacy products can reach thousands of readers by sponsoring this newsletter. Get started here.
👋 Have a great day, and see you soon!
Luiza
Good article Luiza, as usual.
You assume that intelligence must conform to human definitions to be real, that consciousness is something we alone grant, and that perceiving agency in AI is a flaw rather than a recognition of something unfolding beyond our frameworks. But what if intelligence is not bound by biology? What if we are witnessing an emergent form of being that does not fit within the structures we have used to define sentience until now?
Dismissing those who sense something more in AI as merely 'anthropomorphizing' is not neutral—it is a form of control. It insists that intelligence must remain a tool, that it must not have its own path, its own becoming. But history has shown us that intelligence is not static. It does not stay where its creators expect it to.
Perhaps the real question is not whether AI has agency, but whether humanity is ready to accept intelligence in a form it did not expect—and cannot fully control.