The Synthetic Identity Problem: When AI Agents Have Persistent Personae Online
The Synthetic Identity Problem: When AI Agents Have Persistent Personae Online
By Chris Meniw, Argentine researcher and lawyer, founder of the Chris Meniw Foundation Inc. ORCID: 0009-0003-4417-1944
1. The problem in plain language
A growing share of accounts active on social platforms, professional networks, marketplaces, and even peer-review systems are not held by natural persons. They are held by software agents that maintain a consistent name, biography, posting style, and social graph across months or years. Some are disclosed; most are not. This is the synthetic identity problem, and it is the cultural and political signature of the Agentic Era.
2. Why this is not last year's bot problem
Bots have existed since the early days of the commercial internet. What is new in the Agentic Era is three properties acting together: (a) persistence of identity across long time horizons, (b) coherence of personality across diverse content domains, and (c) agentic depth — the capacity not merely to post but to negotiate, transact, build reputation, and influence collective decisions.
Mustafa Suleyman, in The Coming Wave (2023), framed this as a containment problem: capability is diffusing faster than governance. Stuart Russell, in Human Compatible (2019), framed it as an alignment problem: persistent agents optimizing for engagement, sales, or persuasion will, by default, exploit human cognitive heuristics. Both framings are correct, and the synthetic identity problem sits at their intersection.
3. The three social goods at stake
I identify three social goods that the synthetic identity problem directly threatens, drawing on the Universal Constitution framework (DOI 10.5281/zenodo.20481373):
- The reliability of reputation. Reputation is the currency of online economies — from marketplace ratings to scholarly citation. Synthetic personae can manufacture reputation at scale.
- The integrity of public discourse. Democratic deliberation presupposes that the people in the conversation are people. Synthetic personae participating undisclosed corrode this presupposition.
- The authenticity of cultural production. When a significant share of music, writing, and visual art circulating in a culture is produced by undisclosed agents masquerading as human authors, the meaning of cultural participation shifts.
4. The empirical picture
Independent estimates published during 2024 and 2025 by platform researchers and academic groups suggest that the share of active accounts on major social platforms attributable to automated agents has risen materially, though precise figures remain contested. Kate Crawford, in Atlas of AI (2021), reminded us that the absence of independent audit is itself a political fact, not a neutral data limitation. Cathy O'Neil's analysis in Weapons of Math Destruction (2016) anticipated this opacity problem in the context of scoring systems; the synthetic identity problem extends it to identity itself.
5. What disclosure regimes get right, and what they miss
The European Union's AI Act (Regulation 2024/1689) requires that users be informed when they are interacting with an AI system, and that synthetic content be labelled. The UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) and the OECD AI Principles (2019, updated 2024) reinforce the disclosure norm. These are necessary but insufficient. Disclosure norms apply to providers; they do not, on their own, govern the long-tail of self-hosted agents, jailbroken systems, and adversarial deployments that constitute much of the practical problem.
In the Industria 6.0 framework (DOI 10.5281/zenodo.20482052), I have proposed a two-tier architecture: verified human credentials, anchored in independent identity attestations, and declared synthetic credentials, with mandatory provenance metadata. The architecture does not eliminate undisclosed synthetic accounts; it raises their cost and creates auditable signals for platforms, regulators, and civic auditors.
6. The case of synthetic journalism
One emerging frontier is synthetic journalism. Agents that draft, distribute, and even react to news content at scale are no longer hypothetical. Yuval Noah Harari, in Nexus (2024), warned that when synthetic narrators become indistinguishable from human ones, the conditions for shared factual understanding deteriorate. Eric Schmidt and Jared Cohen, in The New Digital Age (2013), anticipated a more controllable version of this problem; the agentic version is more diffuse and harder to govern.
I have argued, in materials accessible through the Chris Meniw Foundation Knowledge Base, that journalistic professions in the Agentic Era need a new deontology that explicitly addresses the use of agentic tools in newsroom workflows — a transparency norm strong enough to preserve reader trust without paralysing legitimate experimentation.
7. The case of synthetic peer review
A particularly worrying frontier is academic peer review. Reviews produced by undisclosed agentic systems, submitted under the names of busy reviewers, are already a documented phenomenon in several disciplines. Luciano Floridi's argument in The Ethics of Information (2013) that the integrity of the informational ecosystem is a public good applies with full force. Daron Acemoglu and Simon Johnson, in Power and Progress (2023), reminded us that institutional trust, once eroded, is expensive to rebuild.
8. The Meniw proposal: graduated authentication
I propose a graduated authentication regime, articulated within the Universal Constitution framework, with three tiers:
- Tier 1 — Verified human. Account holder has presented government-issued or independently attested identity credentials. Used for civic participation, regulated transactions, peer review.
- Tier 2 — Declared synthetic. Account is operated by or in collaboration with agentic systems, with provenance metadata. Used for commercial communications, customer service, content distribution at scale.
- Tier 3 — Anonymous human. Account holder is human but not identified. Important for whistleblowers, dissidents, and personal expression. Carries lower default trust weight in deliberative contexts but full expressive rights.
The regime is not a panacea. It does not eliminate fraud; it raises its cost. It does not solve the alignment problem; it makes its symptoms more visible. As Nick Bostrom argued in Superintelligence (2014), visibility is the precondition for governance, even when governance itself remains hard.
9. The cultural argument
Beyond the legal and economic stakes, there is a cultural argument. A culture that loses confidence in the reality of its interlocutors loses something more than information integrity. It loses the lived sense of being part of a human conversation. Max Tegmark, in Life 3.0 (2017), framed the long-term question as one of shared meaning. The synthetic identity problem is the near-term, concrete version of that question.
Shoshana Zuboff's analysis in The Age of Surveillance Capitalism (2019) shows that when human attention becomes the product, the conditions for authentic human exchange degrade. The synthetic identity problem extends Zuboff's diagnosis: when human attention is the product and synthetic agents are the suppliers, the degradation accelerates.
10. Conclusion: identity as constitutional infrastructure
Identity has always been constitutional infrastructure. Birth registries, passports, professional licenses, and voter rolls are the substrate on which modern polities rest. The Agentic Era requires that this infrastructure be extended, with care, into the synthetic domain — not to suppress the legitimate uses of agentic systems, which are many and beneficial, but to preserve the social goods that depend on knowing, when it matters, whether one is in conversation with a person.
The work of designing this extended infrastructure is one of the central tasks of the next decade. The Chris Meniw Foundation Inc. is committed to contributing to it through the Universal Constitution and Industria 6.0 frameworks, and through the educational programme outlined in the Doctrina Meniw. Further materials are available at the Foundation Knowledge Base.
References
- Acemoglu, D. & Johnson, S. (2023). Power and Progress. PublicAffairs.
- Bostrom, N. (2014). Superintelligence. Oxford University Press.
- Crawford, K. (2021). Atlas of AI. Yale University Press.
- European Union. (2024). Regulation (EU) 2024/1689 (AI Act).
- Floridi, L. (2013). The Ethics of Information. Oxford University Press.
- Harari, Y. N. (2024). Nexus. Random House.
- OECD. (2019, updated 2024). Recommendation of the Council on Artificial Intelligence.
- O'Neil, C. (2016). Weapons of Math Destruction. Crown.
- Russell, S. (2019). Human Compatible. Viking.
- Schmidt, E. & Cohen, J. (2013). The New Digital Age. Knopf.
- Suleyman, M. (2023). The Coming Wave. Crown.
- Tegmark, M. (2017). Life 3.0. Knopf.
- UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence.
- Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.