An AI does not need to be conscious to become a digital actor
Can an artificial intelligence become conscious? The question has fascinated philosophers, neuroscientists, and computer scientists since long before the emergence of large language models. It takes on a new dimension today because artificial intelligence systems speak, see, hear, reason about certain situations, use tools, and are beginning to act with relative autonomy in digital and physical environments.
There is a strong temptation to interpret these capabilities through our own human experience. When a system explains that it made a mistake, changes its approach, takes its previous actions into account, or appears to know some of its limits, we can easily attribute a form of self-awareness to it. Yet the ability to represent information about its own functioning, to detect an error, or to adapt its behavior does not demonstrate the existence of a subjective experience.
Science itself remains very cautious. Researchers still do not have a universally accepted theory that fully explains how consciousness arises in human beings, which makes it even harder to determine criteria for identifying it in a machine. Interdisciplinary work involving, among others, Yoshua Bengio proposed examining various indicators drawn from the main scientific theories of consciousness. Their analysis did not conclude that the AI systems studied were conscious, while emphasizing that no obvious technical barrier would necessarily prevent future systems from satisfying some of those indicators.
For organizations, however, another question becomes far more immediate. An artificial intelligence does not need to be conscious in order to have an identity, a memory, permissions, objectives, and a capacity to act. It does not need to feel its own existence in order to modify data, trigger a transaction, communicate with other agents, control a piece of equipment, or make a series of decisions with real consequences.
The decisive transition of the coming years may therefore be less the one leading from artificial intelligence to artificial consciousness than the one progressively transforming the software we use into a digital actor to which we entrust the ability to act.
The distinction between intelligence and consciousness is a good starting point. A system can possess impressive cognitive capabilities without our being able to conclude that it undergoes a subjective experience. This question recalls the famous Chinese room argument put forward by philosopher John Searle in 1980. In that thought experiment, a person who does not understand Chinese applies rules that allow them to produce perfectly coherent answers in that language. To the outside observer, the system appears to understand; for Searle, the correct manipulation of symbols is not enough to establish the existence of conscious understanding. The philosophical discussion remains open, notably because proponents of functionalist approaches reply that the relevant property could emerge from the system as a whole rather than from each of its components taken separately.
Contemporary systems make this debate even more complex. They no longer process text alone. Multimodal architectures can integrate images, sounds, video, and other information from their environment. Connected to sensors, Edge systems, or robots, they can also use physical data in real time. This multimodality considerably increases their functional perception capabilities without, however, resolving the question of subjective experience. Processing the temperature of an engine, identifying a person in an image, or recognizing a voice demonstrates an ability to process and interpret signals; it does not allow us to conclude that the system experiences anything when it performs those operations. The content behind this reflection raises precisely this difference between sensory processing and lived experience.
This distinction is essential, because it allows us to avoid an error of perspective. We can leave the still-open question of artificial consciousness to neuroscientists and philosophers while recognizing a far more observable technological transformation: the emergence of systems with increasing operational agency.
It can be useful to distinguish three conceptual levels: computational intelligence, operational agency, and possible consciousness. This classification does not claim to constitute a new scientific taxonomy; rather, it helps clarify three phenomena that everyday language tends to blend together. Computational intelligence refers to the ability to process information, to generate, classify, predict, solve certain problems, or develop lines of reasoning. Operational agency appears when a system can use those capabilities to pursue objectives, interact with an environment, choose actions, and adapt its behavior. Possible consciousness would introduce a different dimension: the existence of a subjective experience, of a felt state, or of a form of sentience. This third level remains scientifically uncertain.
For businesses and organizations, the second level is already enough to profoundly transform IT architecture.
From software to digital actor
Traditional software generally behaves according to its code and the events it responds to. An application receives a request, applies rules, and produces a result. A service account has permissions, but it does not normally reflect on the best way to use those permissions to achieve an objective. An API exposes functions, but it does not spontaneously decide to call three others because its first attempt failed.
An artificial intelligence agent can progressively combine several of these capabilities. It can receive a mission, interpret its context, establish a plan, consult data, call tools, communicate with other agents, observe that an approach is not working, try another one, and pursue its objective over an extended period. When persistent memory is added, its present behavior can also be influenced by its previous interactions.
We then begin to see something appear within the information system that deserves to be considered a persistent digital actor. This expression implies no legal personality and no consciousness. It describes an architectural reality: an identifiable software entity with enough continuity, context, autonomy, and capacity to act that its behavior must be governed like that of an actor in the system rather than that of a simple passive tool.
This evolution leads to an even more interesting notion: that of operational existence. A machine does not need to feel that it exists in order to exist functionally within an architecture. If it has a unique identity, a memory, an activity history, relationships with other systems, privileges, functional responsibilities, and a capacity to act, its existence becomes important for the organization's security and operations. This operational existence considerably changes the questions architects must ask themselves.
Identity becomes fundamental
IT has long distinguished between human users, devices, applications, and service accounts. The arrival of agents makes these categories far less sufficient. When an agent acts across several systems, uses several tools, and communicates with other agents, its identity can no longer be reduced to a simple shared API key or to the account of the human user who initially triggered it.
It becomes necessary to know which agent is acting, who authorized it, what mission it has been given, what resources it can use, what data it can consult, and how long its privileges remain valid. Identity must also make it possible to reconstruct the actions carried out and to distinguish those of the agent from those of humans and other systems taking part in the same process.
IAM applied to agents will therefore have to go much further than traditional authentication. The classic question “Who are you and what do you have access to?” will progressively have to be enriched by other questions: in what role are you currently acting, what objective are you pursuing, who gave you this mission, what actions can you perform in this specific context, and are the conditions justifying this trust still in place?
This evolution connects directly with the principle of Continuous Trust. An authenticated identity should not receive permanent trust simply because it was recognized at the start of a session. As an agent accumulates context, calls tools, receives new information, and progresses through a chain of actions, the trust granted can be reassessed according to its behavior, its mission, its environment, and the risk level of the operations it wishes to perform.
AI Onboarding: integrating an agent before entrusting it with work
This vision also gives new depth to the concept of AI Onboarding. Before asking an AI agent to work for the company, it must be taught how to work within the company. This means defining its role, the knowledge it can access, its identity, its permissions, its level of autonomy, its rules for escalating to a human, the expected results, and the conditions for suspending or ending its activity.
The analogy with onboarding an employee obviously has its limits: an agent is not an employee, and we must not attribute to it the human characteristics that come with that status. The analogy nonetheless remains very useful from a governance standpoint. We would not grant a new person permanent access to every system simply because they might one day need it. We should apply comparable rigor to an agent capable of using those systems at a speed far beyond that of a human.
The life cycle then becomes essential. An agent can be created, configured, tested, authorized, deployed, monitored, modified, suspended, and eventually retired. Its permissions must evolve with its role. Its knowledge may need to be updated. Its memory may contain information that must be retained, transferred, or deleted. Its credentials must be revocable when it is no longer in use. Otherwise, organizations risk reproducing the problem of orphaned accounts at scale, this time with entities capable of acting.
Accountability cannot be delegated to the machine
Operational agency then raises a fundamental question of accountability. A system can make a functional decision without being able to assume moral, legal, or organizational responsibility for it. This distinction becomes crucial as autonomy increases.
An agent can determine that a transaction meets the criteria it has been given, choose a supplier, modify a configuration, or recommend a decision. The organization must nonetheless determine in advance which decisions can be executed automatically, which require human confirmation, and which must remain entirely under human authority.
Autonomy thus becomes a delegation of capability rather than a transfer of responsibility. The greater the capability delegated, the more robust the mechanisms for identity, authorization, logging, observability, supervision, and revocation must be.
This logic becomes particularly important when several agents collaborate. A decision may be prepared by one agent, enriched by a second, validated by a third, and executed by a fourth. Without an appropriate architecture, it quickly becomes difficult to determine the origin of a recommendation, the data used, the permissions exercised, and the human responsibility associated with the chain of actions.
When artificial intelligence acquires eyes, ears, and hands
The next transformation appears when these digital actors progressively move beyond purely software environments. Multimodal models already give them access to several forms of perception. Edge AI brings their processing closer to devices and physical environments, while robotics gives them a capacity for material action.
An agent can then observe a camera, interpret sensor data, hear an instruction, monitor a machine, and control a piece of equipment. Whether it “sees,” “hears,” or “feels” in the phenomenological sense remains entirely open. From an operational standpoint, it nonetheless has channels of perception and action that can produce physical consequences.
Research published in 2026 goes even further in drawing on biological inspiration. Researchers are exploring, for instance, so-called interoceptive AI architectures, which mathematically represent certain internal states of the system in order to improve agents' autonomy and their adaptation to changes in their environment. This involves abstracting mechanisms inspired by the homeostasis of living organisms, without that constituting proof of sentience or artificial consciousness.
At an even more experimental frontier are systems combining artificial intelligence and neural organoids. This research must be approached with great caution. An analysis published in 2026 rightly emphasizes that speculation about the possible consciousness of these systems currently goes beyond the available empirical data and risks diverting attention from far more concrete problems: data governance, security, bias, control of closed-loop systems, transparency, and auditability.
This scientific caution strengthens rather than weakens our argument. We do not need to speculate about the imminent birth of artificial consciousness to observe that the boundary between computation, perception, and action is progressively becoming more complex. The evolution that already concerns us can be represented as a technological continuum: AI model, agent, multimodal agent, Edge AI, robotic system, autonomous system, and, over the longer term, various possible forms of biohybrid interfaces. At each stage, the capacity to act increases and the surface to be governed expands.
From cybersecurity to Hypersecurity for artificial actors
This evolution profoundly changes security. When artificial intelligence was used mainly to generate content or perform analyses, it was still possible to treat the model as one application component among others. When an agent has an identity, permissions, a memory, tools, planning capability, and the means to act on the digital or physical world, it becomes necessary to secure its entire operational space.
Traditional cybersecurity principles remain essential: IAM, Zero Trust, segmentation, encryption, logging, access control, monitoring, and incident response. Agency, however, requires orchestrating them together with AI governance, knowledge provenance, decision observability, autonomy management, resilience, and the human capacity to intervene.
This is precisely the territory of Hypersecurity. The question is no longer solely how to protect a server, an account, or a piece of data. What must be protected is an environment in which humans and artificial actors perceive information, use knowledge, communicate, make decisions, and trigger actions.
An agent does not need to feel responsibility for its actions to be traceable. It does not need to morally understand the principle of least privilege for us to have to apply it. It does not need to fear disappearing for us to need a mechanism allowing us to suspend its activity, revoke its identity, or interrupt a chain of actions that has become dangerous. And above all, it does not need to be conscious to produce real consequences.
This distinction should become one of the foundations of governance for agentic systems. Waiting to find out whether a machine possesses consciousness before determining how to govern its autonomy would mean making a current engineering problem dependent on a scientific question that could remain open for a very long time to come.
A strange reversal: AI is already helping us understand our own consciousness
There is, finally, a remarkable paradox in this discussion. While we ask ourselves whether artificial intelligence might one day become conscious, we are already beginning to use artificial intelligence to understand human consciousness.
In March 2026, researchers published work in Nature Neuroscience using an artificial intelligence architecture on more than 680,000 neuroelectrophysiological samples from human and animal contexts in order to study the mechanisms associated with disorders of consciousness. The system made it possible to generate biologically realistic models, to recover certain already-known phenomena, and to produce new predictions that were then tested against various experimental data. Researchers now emphasize that these approaches could contribute to research, diagnosis, and eventually the treatment of disorders of consciousness.
We are therefore using systems whose consciousness we do not consider scientifically established to help us explore the mechanisms of our own consciousness. This situation perfectly illustrates the need to distinguish capability from subjective experience. A machine does not need to experience consciousness to become extraordinarily useful in studying it, just as it does not need to feel a decision in order to take part in preparing or executing it.
The question of artificial consciousness will probably remain one of the great scientific and philosophical debates of the coming decades. We must continue to study it with openness, rigor, and caution, particularly because our understanding of human consciousness itself remains incomplete. The impressive capabilities of current systems are not enough to demonstrate subjective experience, and the current absence of proof does not settle the question definitively for all future architectures either.
For organizations, a far more concrete transformation is already under way. Artificial intelligence is progressively moving from the status of a tool capable of producing an answer to that of a digital actor that can receive a mission, hold an identity, use a memory, access knowledge, hold permissions, collaborate with other agents, and perform actions with digital or physical consequences.
This operational agency demands a new discipline of architecture and governance. AI Onboarding must define the agent's role and limits before it enters service. IAM must assign it a distinct identity and permissions suited to its context. Zero Trust and Continuous Trust must make it possible to continually reassess the legitimacy of its actions. AI governance must define its autonomy and its escalation mechanisms. Cybersecurity and Hypersecurity must ensure traceability, segmentation, observability, resilience, and the capacity for human intervention throughout its life cycle.
At Quantum Beyond, this evolution reflects a fundamental conviction: technological capabilities must progress together with the architectures that make it possible to govern them. Our expertise in AI Governance, AI Onboarding, IAM, Zero Trust, QKS – AI Readiness & Knowledge Governance, Edge AI, cybersecurity, resilience, and Hypersecurity makes it possible to treat the artificial agent as a complete component of the digital ecosystem, with its identity, its knowledge, its access, its context, and its level of authority.
We do not yet know whether a machine will one day truly be able to ask itself: “Who am I?” For architects, AI specialists, and the organizations beginning to entrust it with the ability to act, another question must already receive a far more precise answer: who is it within our architecture, what can it do, who gave it that authority, and how can we regain control? We do not need to wait for machines to become conscious to begin governing them as actors.