Knowledge/A-03ASE

What Is Fail-Closed Behavior?

Governance Review: ApprovedClaims Reviewed: 8Unsupported: 02026-06-08

Fail-closed behavior is a safety posture where, when required inputs, policy context, or evidence are missing or unverifiable, the system refuses to proceed and returns a deterministic, auditable safe response.

Definition

What fail-closed means in practice

Fail-closed means that the system treats uncertainty, missing authorizations, missing evidence, or out-of-band failures as blockers: the safe action is to halt or return a controlled error rather than continuing with a potentially unsafe action.

Why it matters

Reduce the attack surface of assumptions

When systems continue operating without key assurances, they create silent failure modes that can lead to privilege escalation, incorrect outputs, or harmful side effects. For systems that affect safety, privacy, or compliance, failing closed preserves a conservative default that favors human review.

Fail-open vs fail-closed (short comparison)

A fail-open system prioritizes continuity: when a control, policy, or evidence path is unavailable, it favors progressing with the best-available information to preserve availability. A fail-closed system prioritizes assurance: it blocks or defers actions until required guarantees (policy, identity, evidence) are established. Neither approach is universally correct — the right choice depends on the consequence of an incorrect action and the organization's risk tolerance.

Implementation patterns

1.Defensive gates

Use short-circuiting gates that explicitly check for required preconditions (policy loaded, identity present, audit record writable). If any gate cannot assert its precondition, the gate returns a "blocked" decision with a deterministic error code.

2.Policy version pinning

Require that policy evaluations reference a specific, versioned policy artifact. If the artifact is missing or corrupted, treat the evaluation as failed rather than applying a fallback policy implicitly.

3.Append-only evidence path

Ensure that every action attempts to write its audit record before performing side effects. If the record write fails or the tamper-evidence subsystem reports an error, abort the action.

4.Graceful degradations

Provide an out-of-band queue that captures requests for human review when evidence cannot be recorded. The automated path must not continue in the meantime.

Examples

Governance example — invoice approval flow

An AI agent is authorized to create draft invoices but not to approve payments. If the approval policy cannot be loaded (or the authorization service is unavailable), a fail-open system might allow the approval to proceed (prioritizing continuity). A fail-closed design blocks the approval step until policy evaluation succeeds or a human reviewer intervenes via a documented, auditable escalation path.

Failure modes and mitigations

Fail-closed introduces availability tradeoffs; mitigate with highly available evidence paths and auditable escalation channels.

Governance considerations

Fail-closed is a conservative default for governance — it preserves a system's auditable boundary by preventing actions when evidence or authority is absent. Governance teams should:

Related articles

ASE — Auditome

ASE implements fail-closed patterns for high-risk actions: policy pinning, evidence-first writes, and auditable escalations.

Audit Record

article_idA-03
version1.0
statusapproved
review_date2026-06-08
claims_reviewed8
unsupported_claims0
sha256PLACEHOLDER-CHEC

This article passed ASE review, claim validation, and evidence review before publication. Claims are dispositioned as supported by cited literature, Auditome design positions, or verifiable logical consequences of stated definitions.