Trust changes while systems act.
Reassess a principal and its authority as credentials, context, task state and upstream grants change.
- Attributable principal
- Valid authority origin
- Current policy and trust
- Enforceable action decision
Incident-scoped authority is valid.
Risk context changes before the next action.
WRITE and REMEDIATE are withheld.
Authentication is a point in time
A successful sign-in or workload authentication answers a question at one moment. An autonomous task may continue for minutes or hours, invoking tools and APIs as circumstances change. Credential theft, compromised posture, unusual behavior or a changed task can alter the risk of the same operation.
Continuous trust means bringing those changes into the action decision. It does not mean assigning one permanent trust score to an agent. Signals need provenance, freshness and relevance to the action being requested.
- Credential and session risk
- Device or workload posture
- Behavior change
- Task and workflow state
- Policy version
- Upstream authority state
Trust change
Signals and authority must meet
SIEM, EDR/XDR, identity systems and agent runtimes can provide context. They detect or report conditions in their own domains; ProofGrid consumes relevant signals when deciding whether an authorized action may proceed. It does not replace detection or endpoint execution systems.
A signal may narrow what an agent can do without erasing all usefulness. If a security agent’s credential is suspected of compromise, high-risk remediation can be denied while read-only investigation remains available. A parent grant revoked mid-task should stop dependent child actions regardless of their local state.
Reassessment
Reassessment at meaningful points
It may be wasteful to recalculate every signal for every harmless read, while a payment, isolation or physical-control change warrants a fresh decision. The relevant enforcement point depends on the action’s consequence and the architecture of the execution system.
The intended model records which context was considered, why it was sufficient and which decision followed. That makes later review possible without implying all integrations or trust signals are shipped connectors today.
Signal quality is part of authorization
A signal’s presence is not enough; a decision must consider where it came from, how recent it is and whether it applies to the principal and operation being evaluated. A device posture report from yesterday should not silently authorize a production change today. A behavior alert about a different workload should not automatically condemn every agent in the same organization.
The architecture should define what happens when a required signal is absent or delayed. For a high-risk write, missing context may mean deny or escalate. For a read-only diagnostic task, a degraded mode may remain safe. These choices should be explicit, tested and visible in decision evidence.
A long-running task crosses trust boundaries
An agent may authenticate at the start of a workflow, wait for data, delegate analysis and return hours later to make a change. In that time its credential may be rotated, its parent task cancelled, the target system moved into a different operating state or a security incident opened. A single authorization at task creation cannot speak for all of those later conditions.
The decision point should distinguish facts that remain stable, such as the original task purpose, from facts that need fresh evaluation, such as revocation and target state. That lets a system avoid needless checks for harmless intermediate work while still placing a current check before a consequential write.
Signal changes should have proportionate effects
A new risk signal need not always shut down every agent capability. An incident may justify removing write authority while allowing evidence gathering. A parent approval may be withdrawn for one target without erasing unrelated read access. The policy should define the safe subset clearly so a DEGRADE decision is enforceable rather than an informal suggestion.
ProofGrid connects signal provenance to authority and action decisions. The signal producers remain responsible for detecting and reporting their own observations; ProofGrid is not positioned as an EDR, SIEM or threat-intelligence replacement.
Four separate facts at an action boundary
Identity establishes the actor. Authority records who delegated the right to cause an outcome. Trust or risk state describes current evidence about that actor, its runtime and dependencies. Policy defines how those facts affect a requested operation. A risk signal does not erase the origin of a valid grant; it may temporarily narrow the effective permission. Conversely, a good posture score cannot manufacture a grant that never existed.
Signals can include credential compromise, endpoint posture, threat intelligence, behavior anomalies, policy changes, location, dependency compromise and operator intervention. They should come from systems that own the observation, with source, time and freshness recorded. ProofGrid uses relevant identity and trust context for authority decisions. Detection remains the responsibility of the systems that own those signals.
DEGRADE is a defined subset
Suppose a security agent has an incident grant for READ, WRITE and REMEDIATE. A credible compromise signal arrives while it is investigating. Policy can suspend writes and remediation while leaving read-only evidence gathering available. That is DEGRADE: an explicit smaller action set enforced at the next protected request. It is neither a blanket ALLOW nor a universal DENY.
A different operation may require a different fallback. If the chain or approval for a production change cannot be verified, that write normally stops. A read of non-sensitive diagnostic data may continue under an independently valid grant. The decision should record why the narrowed set was chosen and which signal version was used.
Signals and revocation have different lifecycles
A posture report can become stale; a parent task can be revoked; a credential can be rotated; an incident can be transferred to another operator. The decision point needs to distinguish temporary risk from loss of originating authority. A revoked parent cannot be restored by a later green posture signal. A valid grant may become usable again after risk clears only if the chain, expiry, scope and policy still hold.
Define how quickly relevant signals and revocations must reach the enforcement boundary, what happens when a source is unavailable and how an operator can investigate the decision. ProofGrid consumes relevant trust signals through deployment-specific integrations; this public model does not imply a native connector or vendor endorsement.
Platform / Next Step