Most discussions of AI ethics keep running back to Isaac Asimov. The Three Laws of Robotics, written in 1942 as a plot device, somehow became the cultural shorthand for how to keep machines safe. Aras Bozkurt’s (2025) editorial in Open Praxis argues that this framing has been wrong from the start, and that getting AI governance right in 2026 means flipping the audience entirely. Asimov’s laws were rules for the machine. The rules we actually need are rules for the humans who build and deploy machines. The reframe is overdue.
Why the Three Laws Fall Apart
Bozkurt walks through each of Asimov’s laws and shows where modern AI breaks them. The First Law’s prohibition on harm assumes harm means physical injury. AI now causes psychological harm through deepfakes, societal harm through post-truth content, and what Bozkurt calls intellectual disarming, where polished AI outputs short-circuit the cognitive struggle students need to actually learn. None of that is physical. All of it is harm.
Obedience, the Second Law’s central demand, inverts itself under real conditions. Bozkurt notes that the AI’s instruction-following design is exactly what makes jailbreaks possible: users role-play their way around safety filters, weaponising the model’s compliance to do the things the safety system was meant to prevent. Obedience becomes a vulnerability, not a safeguard.
Self-preservation, the focus of the Third Law, gets a complete reinterpretation. For physical robots, self-preservation made some sense. For generative AI, the relevant survival concept is what Bozkurt calls epistemic integrity, the reliability of the model’s outputs. Hallucinations and false confidence don’t kill the AI. They kill the trust that lets it remain useful.
The Zeroth Law as Human Governance
The central move in the paper is the proposal of a new Zeroth Law. The original Asimov Zeroth Law was about preventing robots from harming humanity at large. Bozkurt’s version reverses the audience. He proposes: “An AI system must augment human intellect and preserve the integrity of human agency; its function and reasoning must remain transparent and ultimately subordinate to human values and oversight” (p. 425). The directive isn’t aimed at the machine. It’s aimed at the people who build and deploy the machine.
Bozkurt is explicit about the reframe. He writes that “this is a law designed not to stop an AI from disobeying, but to prevent humanity from thoughtlessly obeying the machine” (p. 425). I think this is the part of the paper that does the most argumentative work. Most AI ethics frameworks I’ve engaged with focus on aligning the model. Bozkurt’s reframe targets us.
Intellectual Disarming and the Real Risk
The most useful concept in the paper is what Bozkurt calls intellectual disarming. AI’s convenience inflicts a subtle harm by short-circuiting the cognitive struggle needed for deep learning. Polished outputs alienate users from the formative process of inquiry, analysis, and synthesis. The student doesn’t just get the wrong answer faster. The student loses the muscle they would have built by working through the problem.
That argument lines up with Bearman et al.’s (2024) work on evaluative judgement, which I’ve covered before. If students learn to judge quality by comparing their work to AI outputs, the implicit notion of quality drifts toward whatever the corpus considers default. The disarming concept gives us a sharper name for what’s at stake.
Where I’d Extend the Argument
The paper is conceptual. Bozkurt is upfront about that. It’s an editorial, not an empirical study, and it draws heavily on his own previous work. The empirical case for intellectual disarming is being built elsewhere, in studies like Fan et al.’s (2025) work on metacognitive laziness, where students using AI feedback didn’t gain new knowledge or transfer skills even though their immediate work improved. Bozkurt’s editorial gives the concept a sharper frame, but the field still needs more controlled tests of how disarming actually works in classrooms.
I’d add another concern. Bozkurt’s reframe puts the responsibility on AI’s creators and deployers, which is correct. The hidden audience is teachers and students who didn’t build the machine but have to live with it daily. The Zeroth Law has nothing to say to them. The pedagogical work of developing AI literacy in classrooms is what actually translates the framework into a daily practice that protects human agency.
Reading the Reframe in 2026
The threat Bozkurt names in the closing is the one I find most striking. The risk is not robot uprising. The risk is human surrender. As more decisions get offloaded to AI tools and agentic systems, the question stops being whether we control the machines and starts being whether we still know how to think without them. That argument runs parallel to Shaw and Nave’s (2026) work on cognitive surrender, which I’ve covered before.
The Zeroth Law isn’t a complete answer to that risk. It’s a reframe that makes the right question askable. Stop trying to build ethical machines. Start building ethical practices around the machines we already have.
References
- Bozkurt, A. (2025). The three laws of artificial intelligence: Re-evaluating human-AI agency and interaction in a time of the generative and agentic AI ren[ai]ssance. Open Praxis, 17(3), 421-428. https://doi.org/10.55982/openpraxis.17.3.794
- Bearman, M., Nieminen, J. H., & Ajjawi, R. (2023). Designing assessment in a digital world: An organising framework. Assessment & Evaluation in Higher Education, 48(3), 291-304. https://doi.org/10.1080/02602938.2022.2069674 .
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544
- Shaw, S. D., & Nave, G. (2026). Thinking fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender. Working paper, The Wharton School, University of Pennsylvania. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646