We build the independent authority and evidence layer for enterprise AI.
FieldHash governs what may shape an AI answer or action, then records the decision for review.
Mission
Keep humans in authority over the AI acting in their name.
Your reviewers, policies, and systems of record establish what an agent may do. FieldHash enforces that authority independently of the model and preserves a packet reviewers can verify under the configured deployment policy. The human remains the author; the model does not police its own boundary.
That is a narrower, more honest promise than “trustworthy AI.”
Long-term vision
Agency does not require sovereignty.
AI systems can reason, act, adapt, and replan without gaining authority merely because they found another path. Customer systems and authorized reviewers remain the source of what may influence the agent and what it may make real.
FieldHash is building toward durable authority that remains explicit, attributable, revocable, and open to authorized challenge as models and runtimes change.
Capability can scale. Authority stays with accountable institutions.
Inspect the current evidenceAuthority Charter
Capability does not create authority.
FieldHash exists to prevent intelligence, convenience, or administrative access from expanding what may affect the world.
Capability does not create authority.
A model, operator, or administrator cannot turn technical reach into permission by itself.
Authority comes from outside the agent.
Customer systems and authorized reviewers define what may proceed. Authority remains attributable, scoped, reviewable, and revocable.
Administrators are governed too.
Exceptional access should be explicit, bounded, time-limited, and evidenced rather than hidden behind an unrestricted override.
Evidence should be minimal.
FieldHash does not require hidden chain-of-thought or treat model confidence as authority. It records the governed handoff and the evidence needed to verify it.
Configured authority is not moral legitimacy.
FieldHash can prove that a decision follows the configured authority state. Customers remain responsible for the legality and legitimacy of that authority.
Safety must preserve useful work.
Success means unauthorized effects stop while authorized objectives continue, with false denial and review burden measured alongside prevention.
From ledger evidence to governed inference
FieldHash grew out of evidence systems reviewers could verify for themselves. We apply that same discipline at the live AI handoff: configured authority determines what may proceed, and the packet preserves why.
What we believe
Govern the path, then keep the proof.
AI needs more than a constitution.
Model self-policing is a soft control. Moving the gate to the handoff enforces configured policy independently of the model, so model choice can turn on capability, latency, and cost without shifting the governance boundary.
Authority is defined, not inferred.
FieldHash does not replace or independently index your systems of record, and it does not guess your rules. You define authority through existing systems and review; FieldHash enforces the resulting signals at the handoff.
Proof matters more than posture.
A narrow claim that can be inspected is more useful than a broad claim the buyer has to trust.
Enterprise AI needs reviewable paths.
When an output is challenged, teams need to know what was allowed in, what stayed out, and why.
What we build
Three surfaces, one governed path.
FieldHash governs configured handoffs across three surfaces: which records may influence an answer, which actions an agent may execute, and which reviewed decisions may govern later cases. FieldHash Ledger preserves the resulting path for review.
Governed Memory
Controls which retrieved records may influence an answer, while retaining stale, rejected, superseded, or rolled-back records for review.
Governed Actions
Controls which tools remain available to the agent, and can require verified human approval before configured consequential actions run.
Governed Precedent
Controls which reviewed decisions may govern future cases: scoped, expiring precedents that suspend on drift and route back to review.
FieldHash Ledger
Records the governed path across all three surfaces: what was allowed, what was blocked, which authority signal applied, and what reviewers can inspect later.
Founder
Aaron Martinez
Aaron founded FieldHash around a narrow operating question: when an AI answer or action is challenged, can the team show which authority was allowed to shape it?
He builds the gate, evidence program, and pilot path around the same rule: publish the boundary, keep the claim narrow, and let reviewers inspect the record.
Write to AaronStart with one workflow.
The fastest evaluation is narrow: one agent or RAG workflow, the authority signals it already depends on, and a shadow-mode comparison before enforcement.
Request a six-week shadow evaluation