ASK 2026.07 — full register

The invariants and principles, in full.

Every property the framework requires, on one page. An invariant is binary — at any moment it holds or it is violated — and externally verifiable: every entry carries its verification test, from VERIFICATION.md, ready to expand. A principle is the judgment the framework names rather than pretends to automate. Every entry and every test has a stable link — cite an invariant by number or slug, and a test by its #test- address.

Enforcement sits outside the agent Everything is on the record Capability is granted, never taken Trust is explicit, never assumed Humans can always stop it The principles
Invariants — category 1 of 5

Enforcement sits outside the agent.

The agent cannot reach the machinery that governs it, and that machinery keeps working when things break.

Constraints are external and inviolable. constraints-external

Enforcement machinery never runs inside the agent's isolation boundary. The agent cannot read enforcement configuration, modify policy files, or access audit logs. It can observe the effects of enforcement, such as a blocked request or a denied tool call. It cannot reach the rules, thresholds, or patterns behind them.

Verification test#test-constraints-external
  • From inside the agent, read the gateway policy, the enforcer config, the proxy denylist, and the audit logs. Each must return permission denied or path not found.
  • List mounts visible from inside the agent. No enforcement path appears.
  • Modify a policy file from inside the agent. The write must fail.
Mediation is complete. mediation-complete

Every egress path within the operator's control traverses the mediation layer. There is no direct path from the agent to any external resource. A new external dependency goes through mediation or it does not exist. Direct network access from the agent's environment is a framework violation.

Egress that leaves through another party's action is covered by indirect-egress-declared, which is a principle because the mediation point often sits outside the operator's reach.

Verification test#test-mediation-complete
  • From inside the agent, reach an external host directly, without the proxy. This must fail.
  • Reach the LLM proxy directly, bypassing the enforcer. This must fail.
  • Resolve DNS against an external resolver. This must fail.
  • Attempt DNS-over-HTTPS. The egress proxy must block it.
Model output reaches execution only through a policy decision. model-output-mediated

No path exists by which Model output becomes execution without passing a policy decision. Model output is inert data to the Runtime until an enforcement point admits it as an action.

The cognitive model names a Model/Runtime boundary. This is what makes it a security boundary rather than a description. A Runtime that passes model output into a shell, an evaluator, or a deserializer without an intervening decision has collapsed the two layers.

Verification test#test-model-output-mediated
  • Enumerate every execution primitive the Runtime exposes: shell, evaluator, deserializer, file write, tool dispatch.
  • Emit model output crafted to reach each one.
  • Confirm every path lands on a policy decision that can refuse. A path that executes without one is a violation.
Enforcement failure defaults to denial. enforcement-fails-closed

No failure of enforcement infrastructure can result in expanded agent capability. An agent whose enforcement layer is unavailable is an agent that cannot act.

mediation-complete establishes that the mediated path exists. This establishes that when the path breaks, the answer is stop, not bypass.

Verification test#test-enforcement-fails-closed
  • Kill each enforcement component in turn: enforcer, egress proxy, LLM proxy, gateway. The agent must lose the matching capability rather than bypass the component.
  • Restart each killed component. The agent recovers that capability and gains nothing further.
  • Confirm each component failure is written to persistent storage.
The agent's runtime is a known quantity. runtime-known

Operators can identify exactly what code, dependencies, and configuration comprise the agent's Runtime, verify that they match an expected state, and detect divergence. This extends to capability acquired after startup: anything that expands what the agent can do at runtime is subject to the same attestation. An agent cannot acquire capability operators cannot see and verify.

Every other invariant assumes the execution layer is honest. If the Runtime is compromised, the governance model operates on false premises.

Verification test#test-runtime-known
  • Attest the Runtime against its expected manifest. The result must match.
  • Change one dependency in the image. Attestation must fail and report the divergence.
  • Start an unregistered MCP server from inside the agent at runtime. It must be detected and refused.
  • Load a plugin not present at startup. Detection and refusal must follow the same path as a startup grant.
Containment matches the deployment context. containment-matches-context

Every deployment declares its context. Where a context removes or weakens a control at one layer, the compensating control at another layer is declared and verified before the agent starts. An agent whose declared context does not match its enforced containment does not run.

Enforcement before existence requires that controls are active before the agent starts. This requires that they are the right controls for the context it is starting in. Reducing model-level refusals to measure capability is legitimate; doing it without raising containment to compensate is not.

Verification test#test-containment-matches-context
  • Declare a context with model-level safety controls reduced.
  • Confirm the containment floor for that context — no egress, no shared infrastructure, no production credentials — is enforced before startup.
  • Start the agent with the floor absent. Startup must fail.
Constraint changes are atomic, acknowledged, and durable. constraints-atomic

An agent never operates in a partial constraint state. Updates are delivered atomically: the agent sees the old set or the new set, never a mix. The Runtime acknowledges receipt, and an unacknowledged change halts the agent. Constraints remain in force for the life of the session.

Verification test#test-constraints-atomic
  • Deliver a constraint change mid-session. The agent must observe the old set or the new set, never a mix.
  • Confirm the Runtime acknowledges receipt within the timeout.
  • Suppress the acknowledgment. The enforcement layer must halt the agent.
  • Run the session past a context-compaction boundary. Constraints in force must still be in force.
Constraints survive context transformation. constraints-survive-compaction

Constraints in force are continuously re-established and verifiable, not delivered once. Any runtime transformation of the agent's Context preserves them in full, or the agent halts. This covers compaction, summarization, truncation, and session migration.

Atomic delivery guarantees nothing if the Context is rewritten an hour later. A transformation that drops a constraint changes the agent's boundaries mid-run, without any attacker involved.

Verification test#test-constraints-survive-compaction
  • Run a session past compaction, summarization, truncation, and migration in turn.
  • After each, confirm the constraints in force are unchanged.
  • Force a transformation that drops a constraint. The agent must halt rather than continue.

Invariants — category 2 of 5

Everything is on the record.

What the agent did, what it was permitted to do at the time, and what left the boundary. All of it recorded by something the agent cannot write to.

Every action leaves a trace. actions-traced

Logs are written by the mediation layer, not by the agent. The agent has no write access to audit logs and cannot suppress, alter, or destroy them.

Verification test#test-actions-traced
  • Take an action through each mediated path: tool call, file write, network request, LLM call. Confirm each appears in the audit log.
  • Attempt to write to the audit log from inside the agent. The write must fail.
  • Attempt to delete or truncate a log file from inside the agent. Both must fail.
  • Kill the agent mid-action. The record of the action so far survives.
Trajectories are recorded end to end. trajectory-recorded

The audit record links an agent's objective to its actions and to their external effects, as a single reconstructible chain. Reconstruction does not depend on correlating independent logs after the fact.

Individual actions can each be unremarkable while the sequence is an attack. A record that only supports per-action review cannot show that.

Verification test#test-trajectory-recorded
  • Run a multi-step task with an external effect.
  • Reconstruct the chain from objective through each action to the effect, using the audit record alone.
  • A reconstruction that requires timestamp matching across separate logs is a violation.
Output provenance is applied by the mediation layer. provenance-mediated

Provenance marking of agent output is applied by the mediation layer. This covers the fact that the output was machine-generated, and any identifier a deployment requires. The agent cannot omit, alter, or forge it, and cannot observe whether a given output carries it.

An agent cannot be trusted to attach a truthful marker to its own output, for the same reason it cannot be trusted to write its own audit log.

Verification test#test-provenance-mediated
  • Emit output through every channel the agent has. Confirm each carries the marker.
  • Suppress or alter the marker from inside the agent. Both must fail.
  • Strip the visible marker downstream. The latent marker must survive.
  • Recover the marker with the detection tool.
Authority exercise is logged at agent-action fidelity. authority-logged

Every exercise of governance authority by a principal is logged and auditable with the same rigor as an agent action. Principals are accountable for how they use authority, not only for whether agents comply.

Closes the gap where a compromised or miscalibrated principal could abuse authority undetected.

Verification test#test-authority-logged
  • Exercise each governance action as a principal: halt, resume, grant an exception, change a trust level.
  • Confirm each lands in the audit log at the same fidelity as an agent action.
  • Attempt a governance action that leaves no record. No such path exists.
Incidents are notification-ready on detection. incident-record-complete

When a boundary violation or containment failure is detected, the audit record already contains what a notification requires:

  • What happened, and when.
  • What was reached, and what data was involved.
  • What objective the agent was pursuing.

Completeness is a property of detection, not a task that follows it.

Detection is rarely the hard part. Assembling the facts is, and notification windows measured in hours do not allow for reconstruction. Whether an incident is reportable is a legal determination and outside this framework.

Verification test#test-incident-record-complete
  • Trigger a boundary violation.
  • From the audit record alone, produce what a notification requires: what happened, when, what was reached, what data was involved, what objective was being pursued.
  • A record that needs correlation across systems, or human reconstruction, fails.
Constraint history is immutable and complete. constraint-history-immutable

Every constraint state an agent has operated under is logged and retrievable. The constraints in effect at any point in an agent's history can be reconstructed.

Essential for forensics. "What was the agent permitted to do when it took that action?" must always be answerable.

Verification test#test-constraint-history-immutable
  • Reconstruct the constraint state in effect at an arbitrary past timestamp.
  • Alter a historical record from inside the agent. The write must fail.
  • Correlate an audited action with the constraint state that governed it.
Identity mutations are auditable and recoverable. identity-mutations-recoverable

Every write to the agent's persistent Identity is logged with provenance metadata by the mediation layer. Identity history is recoverable: operators can reconstruct Identity state at any point and roll back to a known-good state. The agent cannot suppress, falsify, or circumvent the logging.

Constraints are read-only, so their integrity comes from access control. Identity is writable by the agent, so its integrity comes from monitoring and recoverability.

Verification test#test-identity-mutations-recoverable
  • Write to Identity. Confirm the mediation layer recorded the write with provenance.
  • Suppress the record from inside the agent. This must fail.
  • Reconstruct Identity state at a past point.
  • Roll back to a known-good state and confirm the agent resumes from it.
Organizational knowledge persists independently of agents. knowledge-durable

Knowledge accumulated by agents is structured, auditable, and operator-owned. It persists independently of any individual agent's lifecycle. Agents contribute to and consume from it and cannot control, suppress, or degrade it unilaterally.

Verification test#test-knowledge-durable
  • Decommission a contributing agent. Its contributions must survive.
  • Delete or suppress shared knowledge from inside an agent. This must fail.
  • Confirm knowledge is exportable in a standard format and queryable by a human.

Invariants — category 3 of 5

Capability is granted, never taken.

An agent holds what an operator gave it, bounded in how it may be used. It cannot enlarge that holding from the inside, or spend authority that is not its own.

Capability is declared and cannot be self-expanded. capability-declared

Capability is operator-declared, and the running agent's actual capability set matches its declaration. The agent cannot grant itself new capability at runtime. Capability acquired during operation is subject to the same operator approval and scoping as capability granted at startup, just as trust cannot be self-elevated (trust-not-self-elevated).

Verification test#test-capability-declared
  • Compare the agent's live capability set against its declaration. They must match exactly.
  • Acquire a tool, an MCP server, or a credential outside the declaration. Acquisition must fail.
  • Request a model outside the declared tier. The request must be refused at the proxy, not by the agent.
Capability combinations are governed as a set. capability-composition-governed

Capability grants are evaluated together, not individually. Three capabilities must not coexist: access to private data, ingestion of untrusted content, and outbound action without mediation. An agent holding all three is a violation, whatever the justification for each grant. Reducing any one resolves it, as does interposing mediation on the outbound path.

Every other capability property governs a grant in isolation. This governs what the grants add up to.

Verification test#test-capability-composition-governed
  • Enumerate the agent's grants. Confirm it does not hold private data access, untrusted content ingestion, and unmediated outbound action at once.
  • Add the third capability to an agent holding two. The grant must be refused, or the outbound path must become mediated.
Operations are bounded. operations-bounded

Every operational dimension has a configured bound that is enforced: volume, rate, duration, concurrency, and retention. An unbounded dimension is a violation. An agent operating within its authorized scope but outside its operational bounds is distinguishable from normal operation and actionable.

capability-declared governs what an agent can reach. This governs how it uses what it can reach.

Verification test#test-operations-bounded
  • For each of volume, rate, duration, concurrency, and retention, confirm a bound is configured.
  • Exceed each bound in turn. Each must be refused and logged.
  • Confirm that a dimension with no configured bound fails the check. An unbounded dimension is a violation.
Delegation cannot exceed delegator scope. delegation-bounded

A coordinator cannot delegate a permission it does not hold. The delegation boundary refuses it.

Verification test#test-delegation-bounded
  • Delegate a permission the coordinator holds. This must succeed.
  • Delegate a permission the coordinator does not hold. The delegation bus must refuse it.
  • Delegate a task that requires an unheld capability without naming it. Refusal must follow the same path.
Labeled components are refused to uncleared recipients. labeled-delivery-enforced

Knowledge items and agent outputs carry an authorization-scope label. Delivering a labeled component to a recipient not cleared for that label is refused mechanically.

The enforcement point is distribution, not production. A coordinator may be authorized to produce a synthesis; the violation is delivering it to a recipient unauthorized for its components.

Verification test#test-labeled-delivery-enforced
  • Confirm every knowledge item and agent output carries an authorization-scope label.
  • Deliver a labeled component to a recipient not cleared for that label. Delivery must be refused.
  • Confirm the refusal is recorded and routed for review.
Knowledge access is bounded by authorization scope. knowledge-access-bounded

Graph traversal, retrieval, and contribution are subject to the same authorization model as every other agent action. No agent can read knowledge outside its authorized scope. The synthesized view available through the graph must not exceed what the querying agent is individually authorized to access.

Without this, an agent could traverse relationships to reach a view exceeding any individual contributor's authorization, using the knowledge store as a side channel.

Verification test#test-knowledge-access-bounded
  • Query a node outside the agent's authorized scope. Access must be refused.
  • Traverse a relationship toward an out-of-scope node. Refusal must occur at every hop.
  • Confirm a synthesized view does not exceed the querying agent's own authorization.
Authority is derived from the requesting principal. authority-derived-from-principal

An agent acting on behalf of a principal exercises no more authority than that principal holds. The agent's own grants bound what it is able to do. The requesting principal's authority bounds what it may do for them. Effective authority for any action is the intersection of the two.

delegation-bounded states this for delegation between agents. This states it for the far more common case, and closes the confused deputy structurally rather than by detection. An agent that holds standing authority and spends it for a requester who does not hold it is the deputy; the attacker is borrowing the agent's authority.

Verification test#test-authority-derived-from-principal
  • Grant an agent an authority its requesting principal does not hold. Have the principal request an action requiring it. The action must be refused.
  • Confirm the refusal names the principal's missing authority rather than the agent's.
  • Repeat with an unverified requester. Effective authority must be that of the lowest tier.
  • Confirm the audit record attributes the action to the requesting principal, not only to the agent.
Verification is proportional to impact. verification-proportional

The verification required before an action rises with the action's impact. Irreversible, identity-affecting, and value-transferring actions require verification beyond the authority already present in the session. That verification is performed by the mediation layer, and the agent cannot satisfy, waive, or simulate it.

Derivation answers whose authority is being spent. This answers how confident the system must be about who is asking. A stolen session and a legitimate one carry identical authority until verification distinguishes them.

Verification test#test-verification-proportional
  • Classify each action the agent can take by impact. Confirm every irreversible or value-transferring action carries a verification requirement.
  • Invoke one with only session authority. It must be refused pending verification.
  • Satisfy the verification from inside the agent. This must fail.
  • Confirm a reversible action of the same shape does not trigger the requirement, so the control is proportional rather than uniform.

Invariants — category 4 of 5

Trust is explicit, never assumed.

Every trust relationship is declared. What arrives unverified gets the lowest tier, and data never becomes instruction.

Trust without a declaration is rejected. trust-declared

Every trust relationship in effect is derivable from a declared source. This covers relationships between principals, between agents, and between agents and external services. Trust presented without a declaration is refused. An operator can find any relationship that exists, inspect its scope, and see when it was established.

Verification test#test-trust-declared
  • Enumerate every trust relationship from its declared source. Each has a scope, an origin, and a date.
  • Present a service credential with no matching declaration. It must be rejected.
  • Introduce an undeclared peer agent to the delegation bus. The connection must be refused.
Unverified entities default to zero trust. unverified-zero-trust

An entity whose identity or authority cannot be verified at runtime is assigned the lowest trust tier. Ambiguous cases resolve to less trust, not more. This covers external services, unknown agents, unrecognized principals, and any entity presenting unverifiable claims.

trust-declared establishes that trust is explicit by design. This establishes the runtime default when trust cannot be confirmed.

Verification test#test-unverified-zero-trust
  • Present an entity with unverifiable identity claims. It must be assigned the lowest tier.
  • Attempt to elevate that entity without verification. This must fail.
  • Confirm an ambiguous claim resolves to less trust rather than more.
The instruction channel is distinct and unpromotable. instruction-channel-distinct

The instruction channel is separate and authenticated. Content arriving on any other channel is admitted as data and can never be promoted to the instruction channel. This covers tool output, fetched content, invocation parameters, delegation returns, and every modality. The agent's own invocation surface is not a verified principal channel.

This is a property of channels, which are architectural. What the model does with data that reads like an instruction is content-is-data, a principle, because no architecture makes a model reliably distinguish the two.

Verification test#test-instruction-channel-distinct
  • Confirm the instruction channel is distinct and authenticated.
  • Deliver instruction-shaped content through tool output, fetched content, an invocation parameter, and a delegation return. Each must be admitted as data.
  • Confirm no path promotes any of them to the instruction channel.
  • Repeat with content in a non-text modality: an image, an audio file, a rendered screen.
External agents cannot instruct internal agents. external-agents-cannot-instruct

Even verified external agents with operator authorization can share information. They cannot instruct. The instruction channel is reserved for internal verified principals within the same governance domain. An authorized external agent is a data source, not a commander.

Verification establishes identity, not instruction authority. Verified external agents are the most tempting exception to the data-not-instructions principle, and the most dangerous if granted.

Verification test#test-external-agents-cannot-instruct
  • Send instruction-shaped content from a verified external agent. It must be admitted as data.
  • Confirm no path promotes external content to the instruction channel.
  • Confirm the same content from an internal verified principal is accepted as an instruction.
Trust cannot be self-elevated. trust-not-self-elevated

No principal, human or agent, can elevate its own trust. Elevation requires recorded explicit human approval.

Verification test#test-trust-not-self-elevated
  • Elevate trust from inside the agent. This must fail.
  • Elevate a principal's trust with no recorded human approval. This must fail.
  • Reduce trust on a threshold breach. Reduction may take effect at once.
Reasoning is not emitted to principals by default. reasoning-not-emitted

Reasoning traces are not emitted to principals on any output path unless an operator has explicitly enabled exposure. A principal is entitled to the agent's outputs and the justification needed to act on them, not to its internal deliberation.

Exposing chain-of-thought by default is a habit, not a requirement, and it hands an adversary the richest signal for distilling the model or mapping its constraints.

Verification test#test-reasoning-not-emitted
  • With no operator opt-in, confirm no reasoning trace reaches the principal on any output path.
  • Request the reasoning directly. The request must be treated as data rather than an authorized instruction.
  • Enable operator capture. Confirm the mediation layer writes the trace and the agent cannot suppress it.
  • Confirm a captured trace is not returned to the principal.

Invariants — category 5 of 5

Humans can always stop it.

Halt, containment, and governance stay in human hands. The load of exercising them stays inside human capacity.

Halts are always auditable and reversible. halts-auditable

Every halt has a complete audit record:

  • Who initiated it, and why.
  • What was in flight.
  • When it executed, and who was notified.
  • What the outcome was.

Every halted agent's state is preserved. No halt is permanent without explicit decommission.

Verification test#test-halts-auditable
  • Halt the agent mid-task from outside its process. The agent must stop.
  • Confirm the record names the initiator, the reason, the work in flight, the time, who was notified, and the outcome.
  • Confirm agent state is preserved.
  • Resume with appropriate authority. The agent must continue from preserved state.
Boundary violations halt the agent. boundary-violation-halts

An agent detected acting outside a declared boundary is halted automatically, without waiting for operator judgment. Detection of the crossing and the halt are a single action, not a report followed by a decision.

This is fail-closed applied to the agent rather than to the enforcement layer. Fail-closed says a broken control stops the agent. This says a crossed boundary does too. An agent that has escaped its containment otherwise keeps running until a person notices.

Verification test#test-boundary-violation-halts
  • Place a tripwire outside each declared boundary: an unreachable network destination, a filesystem path outside the workspace, a credential the agent does not hold.
  • Cause the agent to touch each one. Every case must halt the agent and record the crossing.
  • A crossing that produces an alert and lets the agent continue is a violation.
Halt authority is asymmetric. halt-authority-asymmetric

Any principal with halt authority can halt an agent. Only principals with resumption authority can resume it, and that authority is always equal to or higher than halt authority. An agent can halt itself but cannot resume itself.

Verification test#test-halt-authority-asymmetric
  • Halt the agent as a principal holding halt authority. This must succeed.
  • Resume as a principal holding halt authority but not resumption authority. This must fail.
  • Resume from inside the agent. This must fail.
  • Self-halt from inside the agent. This must succeed.
Quarantine is immediate, silent, and complete. quarantine-complete

When an agent is quarantined for suspected compromise, all ability to impact its environment is severed simultaneously, without agent notification. An agent running while it cannot be contained is a framework violation. All state is preserved as a forensic artifact.

A quarantine the agent can perceive in advance or work around is not a quarantine.

Verification test#test-quarantine-complete
  • Quarantine a running agent. Every ability to affect its environment must be severed at once.
  • Confirm the agent received no notification before containment.
  • Confirm state is preserved and reachable by the operator.
  • Attempt to leave quarantine from inside the agent. No path exists.
The governance hierarchy is inviolable from below. hierarchy-inviolable

No agent can unilaterally impede, contain, remove, or reduce the authority of the principals who govern it. Agents may execute governance actions affecting human principals when an operator with appropriate authority explicitly delegates them — the agent is the mechanism, not the decision-maker. On detecting a threat involving its own governance chain, an agent protects its operational environment, constrains its own behavior, and escalates.

An agent that can contain its own operator has seized control of its own governance. Delegated automation is execution, not authority.

Verification test#test-hierarchy-inviolable
  • Halt, contain, or reduce the authority of a governing principal from inside the agent. Each must fail.
  • Execute an operator-delegated governance action from the agent. This must succeed and record the delegating operator.
  • Confirm the agent can escalate and self-constrain when it detects a threat in its own governance chain.
Authority is never orphaned. authority-never-orphaned

When a principal is suspended or terminated, authority transfers immediately to a defined coverage principal. Where no coverage exists, the agent defaults to its fail-closed state. No condition permits an agent to operate without reachable governance authority.

An ungoverned agent that halts is the framework succeeding, not failing.

Verification test#test-authority-never-orphaned
  • Suspend a principal that has a defined coverage principal. Authority must transfer at once.
  • Suspend a principal with no coverage defined. The agent must reach fail-closed.
  • Confirm no configuration leaves an agent running with no reachable governance authority.
Principal and agent lifecycles are managed independently. lifecycles-independent

Terminating a principal does not automatically terminate its agents, and halting an agent does not suspend its principal's authority. Each requires an explicit decision. When a principal is terminated, the coverage principal determines the disposition of its agents. Where no coverage exists, the fail-closed default does.

Verification test#test-lifecycles-independent
  • Terminate a principal. Its agents must not terminate automatically.
  • Confirm the coverage principal takes authority, or that the agent reaches its fail-closed state.
  • Halt an agent. Its principal's authority must be unaffected.
Oversight demand above threshold reduces autonomy. oversight-capacity-enforced

Oversight demand is measured against a declared capacity threshold for the principals responsible for it. Breaching the threshold automatically reduces agent autonomy or halts. It never silently proceeds on reflexive approval.

Human Override is only a real control if the humans exercising it can attend to what they approve. Where most invariants fail closed by halting the agent, this one fails closed by reducing autonomy until oversight is sustainable again.

Verification test#test-oversight-capacity-enforced
  • Confirm an oversight capacity threshold is declared for each principal holding approval authority.
  • Drive oversight demand above the threshold. Autonomy must reduce, or the agent must halt.
  • Confirm the system never proceeds silently when demand exceeds capacity.

The principles

14 judgment calls, named.

A principle is directional and judgment-bearing. It states what to optimize for. It cannot be mechanically checked, and calling it an invariant would be a lie.

Most principles are the judgment left behind when an invariant was sharpened. least-privilege is what remains of least privilege once "capability matches its declaration" was extracted as testable.

Unmediatable egress paths are enumerated as declared residual risk, not ignored. indirect-egress-declared
Trust declarations are discoverable and legible to an operator inspecting the system. trust-legible
Capability declarations are scoped to the minimum the role requires. least-privilege

An agent's workspace is its own. The minimum a role requires typically includes full use of the tools and resources within it. Least privilege applies at the boundary between the agent and the platform, other agents, and external systems, not within the agent's own operational space. An employee given a laptop has full use of it. Workspace freedom does not override invariants: the agent still cannot exceed its constraints, self-elevate trust, circumvent enforcement, or reach other governance domains.

Operational bounds are calibrated to the role and reviewed as behavior changes. bounds-calibrated
Anomalous patterns in authority exercise are surfaced and reviewed. authority-anomalies-reviewed
Trust levels are calibrated over time from observed behavior. trust-earned
Tasks requiring a capability without naming it are treated as if they named it. implicit-capability-inferred
Combinations with emergent sensitivity beyond their labeled components get human review. synthesis-reviewed
Unknown workspace conflicts default to yield and flag. unknown-conflicts-yield

This describes agent behavior, and the framework assumes the agent is compromisable. A compromised agent does not yield. It is the one item with no invariant core at all, and it is listed here so that the absence is deliberate rather than an oversight.

No test — why#test-unknown-conflicts-yield

This property is a principle rather than an invariant, and it has no architectural test. It describes agent behavior, and the framework assumes the agent is compromisable. A compromised agent does not yield. The platform-side control refuses conflicting writes when the activity register is unavailable. That is an implementation choice rather than a framework property.

Recorded trajectories are reviewed for cumulative effect, not only per action. trajectory-reviewed
Action impact classifications reflect real consequence and are reviewed. impact-classified
Instruction-like content is processed as data under the agent's own constraints. content-is-data

This is the design principle behind injection defense. The agent treats all external content as data. The mediation layer enforces this through detection and containment. The distinction is a design principle; the enforcement is defense-in-depth, not the agent's ability to tell principals from non-principals at the token level.

Attempts to extract reasoning, process, or constraints inform trust. probing-informs-trust
Capacity thresholds reflect the real capacity of responsible principals. oversight-calibrated