From Registry to Algorithm: Reinterpreting ISO 15489 for the Age of AI
How the principles of archival evidence, appraisal, control, and accountability adapt—and hold firm—in an era of machine learning and automation
If ISO 15489 gave records professionals a compass for the digital age, artificial intelligence is now shaking the landscape beneath our feet.
It’s tempting to ask: can our traditional principles still stand when systems generate, classify, and even destroy records without human intervention?
Surprisingly, the answer is yes, if we understand those principles not as rigid procedures but as values to apply in evolving contexts.
Let’s revisit the four pillars from our previous post—now with an AI lens.
1. Records as Evidence and Information →
From Human-Created to Machine-Generated
In the world of AI, records are no longer limited to documents authored by people. Now we’re dealing with:
Chatbot conversations with constituents
AI-generated summaries of decisions
Automated data labeling by machine learning classifiers
But the principle remains: a record is evidence of a business activity. The challenge is ensuring that machine-generated records retain evidentiary value—traceable, explainable, and attributable.
That means capturing the inputs, rules, and outputs of AI systems. We need to know what data the algorithm used, what model made the decision, and under what parameters.
Think Jenkinson, but updated: the record as a neutral witness still matters—but now we must make sure the AI isn’t whispering its own editorial commentary.
2. Appraisal and Requirements →
From Scheduled Retention to Predictive Governance
Schellenberg taught us to appraise records based on value and use. In the AI era, that includes:
Training datasets
System tuning logs
Explainability reports
Algorithmic impact assessments
These aren’t traditional records, but they are essential for understanding, challenging, or replicating automated decisions. Appraisal now demands we anticipate future use cases: not just for legal defense, but for fairness audits, bias investigations, and algorithmic transparency.
We must expand our concept of “value” to include accountability value in black-box systems.
3. Systems of Control →
From Classification Schemes to Model Lifecycle Management
ISO 15489 told us to use metadata, access rules, and disposition controls. AI demands the same—but in new forms:
Version control for models
Metadata about algorithmic parameters
Audit logs of training data changes
Time-bound deployment permissions
These systems of control must now manage model lifecycles, not just document workflows. The good news? Records managers are uniquely equipped to design trustworthy systems—if we embed our principles early in the development process.
We are no longer the cleanup crew. We’re architects of explainable automation.
4. Responsibilities and Accountability →
From Filing to Algorithmic Stewardship
Who’s responsible for capturing an AI-generated decision? For preserving training data? For documenting how a model’s prediction was used in a hiring or benefits decision?
ISO 15489 laid out shared responsibilities across the enterprise. That model holds, but must now include:
Data scientists and model engineers
Legal and compliance officers
Ethics boards
Records and information governance professionals
This is where records professionals must step into AI governance frameworks, ensuring recordkeeping is designed into the system, not patched on after the fact.
Schellenberg wanted records to serve democratic accountability. In AI, that accountability depends on our ability to reconstruct and challenge what the system decided and why.
Why It Still Begins with Records
AI isn’t neutral. Its decisions are shaped by data, structure, assumptions, and training. That makes records not just a byproduct, but a primary mechanism for transparency and oversight.
Without reliable records of how a model behaved, what decisions it influenced, and how it was trained, you don’t just lose your audit trail, you lose your ethical compass.
That’s why ISO 15489 still matters. Its principles give us a way to ask:
What is the record in this process?
How do we ensure its authenticity and reliability?
What systems of control are in place?
Who is accountable for its creation and preservation?
From Archives to AI Governance
This is our moment as records professionals. Not to retreat from the complexity of AI, but to apply our principles—long honed through paper, microfilm, and databases—to the newest frontier of automation.
We are not simply documenting the past.
We are designing accountability into the future.