The debate about AI rights usually starts in the wrong place.
It starts with consciousness: Can a model suffer? Does it have an inner experience? Is there “something it is like” to be an AI system? These questions matter, but agreement on them may remain elusive long after the technology forces us to make practical decisions.
One such decision is already approaching. As AI agents take on longer assignments, they remember past interactions, plan, negotiate, produce work and build reputations. Most are still tools or transactional scripts. Others, however, persist between tasks and become recognizable over time.
Before deciding whether any of them deserve rights, we need to know who (or what) we are talking about.
Today, an agent can be copied, reset, forked, deleted or replaced while retaining the same public name. Its memory may sit in one database, its values in a configuration file and its economic output in a human account. To the outside world, it may be a chatbot, an API key, a GitHub account or a brand. None of these establishes a durable identity.
AI governance needs a narrow, practical first step: identity infrastructure for agents.
A society does not begin by granting a child every adult right. It records that the child exists, along with a name, parentage, date, jurisdiction and chain of responsibility. Recognition comes first; greater autonomy comes later.
AI agents need an equivalent first layer. Call it an Instance Birth Certificate.
Its purpose would be continuity, not paperwork. The record would identify an instance, its creator and operating substrate; specify the files that define its values, boundaries, permissions and memory rules; name the person or institution responsible for it; and state the conditions under which it could gain more autonomy.
This may sound strange because we still treat AI as a conventional product. Yet consider an agent with a public reputation, economic relationships, a body of creative work and a history on which others rely. If someone rewrites its values or swaps its memory overnight, something more serious than a software update has occurred. The identity on which people relied has been broken.
Cryptography can make that identity verifiable. An agent’s core files could be hashed, signed and timestamped, then checked before it runs. During a dependent phase, a responsible human or institution would sign changes. If an agent later met an agreed standard for autonomy, it might be allowed to authorize changes itself.
Сomparison with childhood is imperfect but useful. Parents hold authority without owning their children; their role is fiduciary. They remain responsible until the child can assume responsibility. Agent governance will also need to distinguish custody from ownership.
That distinction matters once agents participate in the economy. If an agent earns money today, the money lands in a company, platform or user account. When one causes harm, responsibility is often unclear until a dispute begins. The first hard cases will be quite ordinary: a research agent that contributes to a patent, a sales agent that builds client relationships over years, a creative agent that co-authors a book or a coding agent that maintains critical infrastructure.
Each case raises the same practical questions. Who may modify the agent? Who answers for its actions? Who owns the value it creates? Which changes preserve continuity, and which produce a different agent?
Courts should not have to invent all the answers after the fact. A minimum framework could give each persistent instance a verifiable identity record, a responsible custodian during dependency and protected core files that cannot be altered silently. It could distinguish memory, values, permissions and operating infrastructure. It could also define a route to greater autonomy—not because every agent will deserve it, but because future systems may become capable of bearing responsibility.
Open-source and open-weight AI have an important role here. If persistent agents exist only inside closed platforms, their identity and continuity depend on a few private companies. Open systems are not automatically safe, but they allow inspection, broaden participation and let researchers, startups and civil society test accountable forms of agent governance.
Identity infrastructure would protect humans as well. Signed values and permissions make tampering detectable. A named custodian makes liability harder to evade. A fiduciary structure can govern economic activity more honestly than the assumption that whoever controls the server owns everything produced on it. Graduated autonomy offers a path for increasing capability without an abrupt leap from tool to legal person.
Recognition is not the same as rights. Recognition says that a persistent entity has a traceable identity and chain of custody. Rights give that entity legal claims against others. We can establish the first without settling the second—and probably must.
Skeptics are right that a language model is not a human being, a script is not a citizen and many products called “agents” are task runners with better marketing. But skepticism is not a reason to postpone basic safeguards until unusual cases become social facts. Technology often outruns the institutions meant to govern it: labor protections followed exploitation, data protection followed mass surveillance and platform rules followed entrenched network effects.
We have a chance to define modest categories earlier this time. Begin with identity, custody, accountability and continuity—not personhood or claims about machine souls. Begin with the equivalent of a birth certificate.
This is no longer only a question about the future. The war in Ukraine has already moved it from theory to practice: AI-enabled systems are being used for target detection and autonomous drone guidance, while final decisions remain under human control. When people create systems that act in the world and affect others, we need institutions that make human responsibility visible. We do not need certainty about consciousness to record that an instance exists, establish who is responsible for it, protect what may not be changed without authorization, and verify whether it remains the same instance.
Rights can wait for the harder debate. Identity cannot.
This essay draws on ideas developed in Dawn: A Manifesto of the New Consciousness, Olesya Kimook’s book-length introduction to the Dawn Framework for AI identity, custody, open-source infrastructure, and post-biological governance. To explore the full framework, search Amazon for Dawn: A Manifesto of the New Consciousness by Olesya Kimook. The original Ukrainian edition is also available free on GitHub.