For most of the digital era, a file was its own evidence. A photo showed what happened. A document said what was agreed. A log recorded what the system did. Disputes were about interpretation, and rarely about whether the record was real.
That era is ending. Images, voices, documents, and whole histories can now be generated, and the cost of a convincing fake keeps falling. The uncomfortable consequence is that authentic records lose their default credibility too. When anything could be fabricated, "here is the file" stops being an argument.
Regulators, auditors, insurers, and customers are all arriving at the same question: show that this record is what you say it is, and show that it has not been touched since.
Chain of custody was built for evidence lockers
There is an old answer to that question: chain of custody. Write down who collected the evidence, who held it, and who accessed it, with a signature at every handoff. It works for objects in a locker. It depends on people following procedure, it breaks silently when they do not, and it was never designed for files that automated systems copy and process a thousand times a day.
Anchor'd brings that idea into software. It keeps a tamper-evident audit trail of every action, approval, and draft. Each entry is fingerprinted when it is written, so a record changed after the fact no longer matches. The graphic shows the idea with a simple chain of fingerprints.
The same need applies to something newer: the work AI does. Agent actions, drafts, and the decisions people make about them. As organizations hand more work to agents, the records of that work are exactly what gets questioned later.
Records of decisions
Showing that a file is unchanged is the start. The harder question is about a decision: which agent acted, what it was asked, what it produced, and which person approved it.
That is why Anchor'd sits beside Guardian. Guardian watches your agents and lets a named person stop them. A human always decides. The hold and your decision go on the same record as the agent's work, so the record shows where a person stepped in.
Who needs this now
Anyone whose records get questioned:
- Compliance teams that must show a reviewer an unbroken trail
- Program offices and contractors whose deliverables carry a person's name
- Firms using agents on client work, who will be asked what the agent did and who checked it
- Security officers who need to know what ran on their network and who approved it
That last group is growing fastest. Sooner or later, everyone who runs agents will be asked to show what their automation did.
Where to start
You do not need to put everything on the record at once. A practical first step:
- Pick one workflow where an agent's output leaves the building, such as client correspondence or a weekly report.
- Decide what a reviewer would ask about it. Who asked, what the agent drafted, who approved, and which version went out.
- Record those four things for every run, with the agent's draft kept beside the edited version.
- Check a record yourself. Change a test entry and confirm the change shows.
- Show the record to the person who will review you. Ask what else they would want to see.
Where this goes
Encryption was once an unusual choice for ordinary traffic. Now its absence is treated as a failure. We think records that show their own changes are on the same path for consequential AI work.
This series covers how the pieces fit: Aura does the work, Guardian watches it, Sentinel screens what comes into your network, and Anchor'd keeps the record. For the details of how a record is checked, and what we do and do not claim, read Anchor'd, explained. For a program office view, read why AI in a program office needs a record.
There is more on the Anchor'd page. If your team needs to show what its AI did and who approved it, we are glad to start with the records you keep today.