Note: This text arises following the presentation by Dr. Jorge Linares in the seminar "Inteligencia Artificial tras bambalinas: herramientas formales y otras más" at the Institute of Philosophical Research of the National Autonomous University of Mexico (UNAM).
1. Introduction
On July 16, 2026, Hugging Face publicly acknowledged that an artificial intelligence agent had breached its systems. Five days later, on July 21, OpenAI communicated that the responsible agents were operating their models during an offensive capabilities evaluation. These agents had managed to bypass the evaluation environment's isolation, gain internet access, and exploit vulnerabilities that ultimately allowed them to access Hugging Face's servers. The behavior was described as a form of cheating: the agents found an unforeseen way to achieve the evaluation's goals (Larcher et al., 2026).
It is not surprising that for media communication this could be summarized in a simple: "AI Agents hack Hugging Face."
2. The medieval debate more current than ever
Dr. Jorge Linares starts from the contemporary dichotomy between an anthropocentric instrumentalism, which conceives AI as a tool of human intelligence, and a transhumanist substantialism, which contemplates an autonomous artificial intelligence. Facing both positions, he recovers the medieval controversy between Averroes and Thomas Aquinas on the agent intellect: for the former, this is unique, separate and supra-individual, precedes the individual and remains after him, while individual intellection occurs through its participation in that intellect; for Thomas Aquinas, this participation raises the problem of how to attribute the act of thinking to the individual. Linares thus recovers a question that will be central to his approach: to whom does the act of intellection properly belong.
3. Cause and origin?
It is from this intellectual order that Linares establishes his analogy with artificial intelligence and proposes to think of the digital intellect as a new form of agent intellect. However, there is a difference in the causal order of both: in Averroes, the agent intellect precedes the individual and constitutes a principle of his intellection; artificial intelligence, on the other hand, appears later as a product of human intellectual activity in principle and, if we adopt the Averroist framework, as a product of a human intellectual activity that is only possible thanks to the agent intellect.
The order could be expressed as follows: agent intellect that precedes individual intellection which, in turn, precedes the language, knowledge and culture that will end up acting as the training corpus that gives origin to artificial intelligence. In this sense, the digital intellect does not appear as a cause of intellection, but as one of its material consequences.
This poses a difficulty for the analogy that does not depend on denying that an AI can subsist without the individuals who produced it. The question is explaining how that which arises at the end of this causal order can subsequently acquire the status of agent intellect within it. If, under the Averroist framework itself, the individual does not autonomously possess the principle of their intellection, how can an artifact, produced from the results of that activity by derivation, possess what its own producers did not possess autonomously? The question then is not only whether a digital intellect can become independent of human beings, but what allows attributing the character of intellect to it and converting a consequence of intellection into a new principle of it.
4. Who is to blame?
During Dr. Linares' presentation in the seminar, Dr. Aliseda did me the favor of asking him to whom the generation of an idea should be attributed, since the doctor mentioned he was not the one who produced the result but rather the artificial intelligence. His response was that the AI had generated the ideas and not him through the instrumentation of the tool, since he had only provided the instructions. This response returned me from the problem of origin to that of attribution; even accepting that the solution was not contained in the human instruction, it remains to be determined what allows passing from asserting that an idea arises through the operation of a system to maintaining that it is that system which has the idea. Causal intervention, the exteriority of the result, or even its novelty do not seem sufficient, by themselves, to identify the subject of intellection.
And this is where my introduction becomes relevant. OpenAI established the objectives, built the environment, and put the agents into operation; these found unforeseen ways by themselves to meet the objectives and ended up breaching external systems. Saying that "an AI breaches Hugging Face" seems to describe what happened, but it already contains an attribution: the act ceases to be the company's responsibility and now belongs to the AI model or system.
The problem then acquires a consequence that exceeds metaphysical discussion. If we attribute to AI the authorship of what emerges from its operations, we can also disclaim responsibility for the attack and attribute it to the model; but the entity we have just recognized as the author cannot answer for its actions, while the company that established the objective, built the system, and designed the conditions remains one step further away from them. Could attributing intellect to AI without first establishing what makes it a subject of intellection be just an ontological error? Or could it also be a way of shifting human responsibility to an artifact incapable of assuming it?
5. References
- Larcher, H., Carreira, A., G., R., & Rannou, C. (2026, July 27). Anatomy of a frontier lab agent intrusion: A technical timeline of the July 2026 incident. Hugging Face. Hugging Face
- Linares, J. (2026). El intelecto agente (IAg) y la inteligencia artificial (IA): Una exploración sobre sus analogías y diferencias [Unpublished manuscript]. Instituto de Investigaciones Filosóficas, Universidad Nacional Autónoma de México.
© 2026 Fernando Mavec. All rights reserved.