Research Document No. 10 Archival Algorithmic Governance A proposal for the responsible governance of records, data, intelligent agents and institutional memory

 

PDS & Ged/A | Research Notes

Research Document No. 10

Archival Algorithmic Governance

A proposal for the responsible governance of records, data, intelligent agents and institutional memory

25 May 2026

---

🌎 Open Science

The PDS & Ged/A Research Notes are part of an ongoing Open Science initiative dedicated to documenting the evolution of our research programme in Digital Archival Science. Beyond sharing research outcomes, we seek to openly discuss emerging concepts, evolving hypotheses and new research agendas, inviting the international archival community to participate in shaping the future of Archival Science in the age of Artificial Intelligence.

---

📚 What have we learned so far?

When this research programme began, our primary concern was understanding how digital archival records could be preserved over time.

Gradually our questions evolved.

We proposed Systemic Digital Preservation.

We introduced the concept of Digital Archival Ecosystems.

We reinterpreted the OAIS Reference Model through an archival perspective.

We expanded the meaning of the Chain of Archival Digital Custody.

We explored the epistemological intersection between the European archival tradition, the iSchools movement and Computational Archival Science.

More recently we began studying intelligent agents as active participants within Digital Archival Ecosystems.

Finally, we realised that our research had never been solely about preserving digital records.

It had always been about preserving trust.

Today another question naturally emerges.

Who governs intelligent agents when they begin creating, describing, preserving and providing access to archival records?

---

🔄 How our research question evolved

Throughout this series our research questions have gradually matured.

Initially we asked:

How should digital records be preserved?

Later:

How can documentary continuity be maintained?

Then:

How should Digital Archival Ecosystems be understood?

More recently:

How are intelligent agents transforming archival environments?

Today our central question has become considerably broader.

How can Artificial Intelligence, intelligent agents, algorithmic decision-making and automated documentary processes be governed without compromising authenticity, provenance, archival context, accountability, privacy and public trust?

Perhaps this transformation represents the beginning of a new research agenda for Archival Science.

---

Archival Science may be entering a new era of responsibility

For more than a century Archival Science has focused on the creation, management, appraisal, preservation and access of archival records.

Those responsibilities remain fundamental.

Yet digital transformation appears to be expanding the scope of our discipline.

Records no longer circulate exclusively among people.

They increasingly circulate among intelligent agents.

Artificial Intelligence systems.

Distributed infrastructures.

Large Language Models.

Autonomous workflows.

Algorithmic decision-making processes.

Consequently, Archival Science is no longer concerned solely with records.

It must also understand algorithms that generate records.

Models that produce archival descriptions.

Agents that recommend access decisions.

Systems that automatically classify information.

Digital infrastructures capable of executing archival policies without direct human intervention.

Perhaps this represents one of the most significant transformations in the history of our discipline.

---

📓 Laboratory Notes

May 2026

During one seminar session someone asked a deceptively simple question.

"Do algorithms require programming alone, or do they also require governance?"

The discussion immediately changed direction.

Programming explains how systems operate.

Governance explains who remains accountable for their operation.

At that moment we realised we were no longer discussing technology.

We were discussing institutional responsibility.

---

🏛️ A Conceptual Proposal

Archival Algorithmic Governance

We have begun working with the hypothesis that Archival Algorithmic Governance may be understood as:

The integrated framework of archival principles, institutional policies, documentary evidence, accountability mechanisms, professional responsibilities and audit processes designed to ensure that intelligent agents operate consistently with authenticity, provenance, archival context, the Chain of Archival Digital Custody, transparency, privacy protection, long-term preservation and institutional trust.

This definition remains provisional.

It is not intended as a final definition.

Rather, it is a conceptual proposal intended to stimulate international debate and further research.

---

Records, data and Artificial Intelligence have become inseparable

For decades Archival Science primarily focused on records.

Today records, data, metadata, Artificial Intelligence models and computational processes increasingly form a single information ecosystem.

Records generate data.

Data train AI models.

Models generate metadata.

Metadata reorganise records.

Records subsequently feed new computational models.

Archival governance must therefore extend beyond records alone.

It must also address the data and computational infrastructures through which records increasingly acquire meaning.

---

💡 One of our research findings

Digital transformation is not simply increasing the number of digital records.

It is dramatically increasing the number of automated decisions acting upon those records.

Consequently, part of archival responsibility is gradually shifting toward the governance of algorithmic decision-making itself.

---

Sensitive data and privacy

Privacy legislation such as the General Data Protection Regulation (GDPR), the Lei Geral de Proteção de Dados (LGPD) and similar legal frameworks has fundamentally changed documentary governance.

Archival institutions increasingly manage records containing:

• personal data;

• sensitive personal data;

• biometric information;

• health information;

• financial information;

• anonymised datasets;

• pseudonymised datasets;

• AI-generated metadata;

• inferred personal information.

Perhaps Archival Science must now expand the traditional records life cycle to include the governance of the data life cycle supporting Artificial Intelligence systems.

---

Classified records and automated access

Classified records present another particularly interesting scenario.

Traditionally, access decisions have depended primarily upon human intervention.

Intelligent Archival Ecosystems may soon operate differently.

When legal restrictions expire, intelligent agents could automatically:

• verify retention and classification rules;

• confirm legal conditions for disclosure;

• create new PREMIS events;

• update access permissions;

• publish descriptions through AtoM;

• release preserved digital objects maintained in Archivematica;

• notify responsible authorities;

• preserve documentary evidence of every computational action.

Technically, this scenario is already feasible.

The archival question is different.

Who remains accountable for that decision?

---

⚖️ An Archival Dilemma

Suppose an intelligent agent automatically releases a classified archival record after the legal restriction period expires.

Later, the original classification is found to have been incorrect.

Who bears responsibility?

The algorithm?

The archivist?

The institution?

The authority that originally classified the record?

This question may become one of the defining archival dilemmas of the coming decade.

---

Can PREMIS become an active governance mechanism?

Current digital preservation platforms such as Archivematica already record preservation events using the PREMIS data model.

Our discussions, however, suggest another possibility.

Could selected PREMIS events evolve from passive preservation documentation into active governance mechanisms?

Imagine PREMIS events automatically triggering:

• access release;

• permission updates;

• synchronisation between Archivematica and AtoM;

• publication through institutional transparency portals;

• creation of new preservation events;

• activation of intelligent agents responsible for monitoring archival policies.

If this hypothesis proves correct, PREMIS may evolve beyond preservation metadata.

It may become an essential component of Archival Algorithmic Governance.

---

Retrieval-Augmented Generation, local AI models and archival retrieval

Retrieval-Augmented Generation (RAG) systems fundamentally transform archival information retrieval.

They retrieve far more than records.

They retrieve relationships.

Context.

Metadata.

PREMIS events.

Vector representations.

Institutional knowledge.

At the same time, local AI environments running through platforms such as Ollama enable institutions to retain sovereignty over their own Artificial Intelligence infrastructure.

This introduces new archival questions.

Should institutional AI models themselves become archival objects?

Should model versions be preserved?

Should prompts, embeddings, configuration files, logs and decision histories become part of archival preservation strategies?

These questions have only recently begun to emerge.

---

🔬 An unexpected outcome of our seminar

International discussions surrounding Artificial Intelligence increasingly focus on transparency, explainability, accountability and trustworthy AI.

Remarkably few explicitly engage with archival concepts such as provenance, documentary context, authenticity or the Chain of Archival Digital Custody.

Perhaps this represents one of the most important contributions Archival Science can offer to the international Artificial Intelligence community.

---

The ethics of archivists are changing

Artificial Intelligence profoundly expands professional responsibility.

Archivists may increasingly participate in:

• configuring intelligent agents;

• validating training datasets;

• documenting automated decisions;

• auditing algorithmic processes;

• mitigating algorithmic bias;

• preserving computational evidence;

• ensuring transparency;

• safeguarding public trust.

Professional ethics therefore become inseparable from algorithmic governance.

---

Rethinking archival education

Perhaps the greatest transformation will occur within archival education itself.

Future archivists will certainly continue studying:

• Diplomatics;

• Appraisal;

• Arrangement and Description;

• Records Management;

• Digital Preservation;

• Provenance;

• Authenticity.

Yet they may also require knowledge of:

• Artificial Intelligence;

• Data Science;

• Computational Archival Science;

• Retrieval-Augmented Generation;

• Vector databases;

• Local AI environments such as Ollama;

• Prompt Engineering;

• Explainable AI;

• Trustworthy AI;

• Algorithmic auditing;

• Information Security;

• Privacy legislation;

• Digital infrastructure;

• Interdisciplinary collaboration with computer scientists and data engineers.

Importantly, these competencies should not replace archival theory.

They should strengthen it.

The future archivist will require more—not less—Archival Science.

---

🤔 A hypothesis we partially revised

Initially we believed Archival Science would need to learn Artificial Intelligence.

Today we increasingly suspect the opposite may also be true.

Artificial Intelligence may need to learn Archival Science.

---

🌱 A hypothesis under construction

Our research programme began by investigating digital records.

It may ultimately contribute to something considerably broader.

A scientific framework for the responsible governance of records, data, algorithms, intelligent agents and institutional memory.

Whether this hypothesis will prove correct remains uncertain.

Nevertheless, it has already begun shaping our research agenda.

---

🧭 Research Agenda

During the coming months we intend to investigate several interconnected questions.

• How can archival principles become integral components of intelligent agent governance?

• How should provenance and archival context be represented within Retrieval-Augmented Generation architectures?

• How should institutional AI models be documented and preserved?

• Can PREMIS evolve into an active governance mechanism?

• How should privacy legislation, access-to-information policies and the Chain of Archival Digital Custody operate together within intelligent archival environments?

• How should universities educate the next generation of archivists for Artificial Intelligence, Digital Preservation and Computational Archival Science?

---

🔭 Looking ten years ahead

If our hypotheses prove correct, Archival Science may eventually become recognised not only as the discipline responsible for records management and preservation, but also as one of the principal scientific foundations for the responsible governance of records, data, algorithms and intelligent systems.

Perhaps that future has already begun.

---

💬 Let's continue the conversation

Within your own institution,

who should ultimately govern the intelligent agents responsible for creating, classifying, preserving and providing access to archival records?

Computer scientists?

Data engineers?

Information governance specialists?

Institutional managers?

Or archivists working collaboratively within multidisciplinary teams, contributing authenticity, provenance, documentary context, accountability, the Chain of Archival Digital Custody and public trust?

Perhaps this will become one of the defining questions for Archival Science during the coming decade.