AI Change
Governance
The checkpoint between what a machine proposes and what runs.
AI change governance is the discipline of deciding, before execution, whether a routed change from a person, pipeline, or AI agent should run: tested on a safe group first, authorized by a named human where risk demands it, with a recovery point confirmed and a tamper-evident record kept. AuthorityGate Keystone is designed to deliver it as a checkpoint the customer owns.
Status: invitation-only early access. Keystone is not generally available. Last updated .
-
8
Configurable gates, from Pre-Validation & Backup to Recovery Readiness
-
Gate 7
SME Approval: a named human, set by policy per change type
-
1
Checkpoint for people, pipelines, vendors, and AI agents
-
Early access
Invitation-only through the Founding Members Program
Change now moves at machine speed
-
AI agents now change infrastructure, configurations, and code at machine speed.
-
Many outages begin with a change, not an attack.
-
Approval tickets show intent. They do not show how the change will behave in your environment.
For survey evidence and primary framework sources, linked with sample and sponsorship context, see the AI Change Governance Field Manual.
The shift in one table
| Dimension | Typical today | With an AI change-governance checkpoint |
|---|---|---|
| Question asked | Was it approved? | Was it tested, who authorized it, and can we recover? |
| Evidence | Ticket, approval email | Gate results, named approver, tested recovery, tamper-evident record |
| Speed | Humans slow the change | Low-risk changes pass on evidence at machine speed; people are interrupted for judgment |
| Recovery | Separate backup process | Recovery readiness is a condition of the decision |
Question asked
- Typical today
- Was it approved?
- With an AI change-governance checkpoint
- Was it tested, who authorized it, and can we recover?
Evidence
- Typical today
- Ticket, approval email
- With an AI change-governance checkpoint
- Gate results, named approver, tested recovery, tamper-evident record
Speed
- Typical today
- Humans slow the change
- With an AI change-governance checkpoint
- Low-risk changes pass on evidence at machine speed; people are interrupted for judgment
Recovery
- Typical today
- Separate backup process
- With an AI change-governance checkpoint
- Recovery readiness is a condition of the decision
Where a change-governance checkpoint fits
Each category below does important work. The question is what it typically does not decide. Keystone is designed to complement these tools; the customer's tools keep delivering change.
| Approach | What it typically does well | What it typically does not decide | Where Keystone is designed to fit |
|---|---|---|---|
| ITSM / CAB | Records intent and approval | Does not test the runtime effect of the change | Adds a tested outcome and recovery evidence to the approval |
| Backup / DR | Restores data | Does not decide whether a change should run | Makes recovery readiness a precondition |
| Security scanners | Find known threats | Do not check compatibility with your live configuration | Security scanning is one gate of eight |
| AI gateways / guardrails | Filter prompts and outputs | Do not decide infrastructure change | Route agent-proposed changes through the same checkpoint |
| Patch / MDM tools | Deliver updates | Do not test against your environment first | A checkpoint before delivery |
ITSM / CAB
- What it typically does well
- Records intent and approval
- What it typically does not decide
- Does not test the runtime effect of the change
- Where Keystone is designed to fit
- Adds a tested outcome and recovery evidence to the approval
Backup / DR
- What it typically does well
- Restores data
- What it typically does not decide
- Does not decide whether a change should run
- Where Keystone is designed to fit
- Makes recovery readiness a precondition
Security scanners
- What it typically does well
- Find known threats
- What it typically does not decide
- Do not check compatibility with your live configuration
- Where Keystone is designed to fit
- Security scanning is one gate of eight
AI gateways / guardrails
- What it typically does well
- Filter prompts and outputs
- What it typically does not decide
- Do not decide infrastructure change
- Where Keystone is designed to fit
- Route agent-proposed changes through the same checkpoint
Patch / MDM tools
- What it typically does well
- Deliver updates
- What it typically does not decide
- Do not test against your environment first
- Where Keystone is designed to fit
- A checkpoint before delivery
Vendor assurance vs. a test in your environment
Most update and change tooling typically releases a change on the vendor's own assurance: release notes, signatures, or reliability scores pooled from other customers. Keystone is designed to test the routed change on a safe group in the customer's own environment first, before authorization.
A vendor can say an update is safe. Only your environment can show it is safe for you.
Vendor assurance is evidence of provenance, not of compatibility with your configuration.
Reliability scores from other customers' systems are a signal. A test on yours is the evidence.
AI agents inherit the same blind spot: they act on what they were told is safe. Keystone is designed to check before it lands.
What Keystone is designed to do
Operational Resilience that keeps up with AI speed: AI Governance and Validation, with human experts in the loop.
Customer-owned, vendor-neutral
Designed to run as a checkpoint the customer owns and configures. The customer's existing tools keep delivering production change.
Eight configurable gates
Pre-Validation & Backup, ITSM Window Check, Zero Trust Verification, Security Scanning, Dependency Health, Behavioral Resilience (BlockStack), SME Approval, and Recovery Readiness, including the customer's own approval. Configured policy decides which gates evaluate, skip, or block.
A tamper-evident record
Decisions are kept in a tamper-evident, hash-chained ledger with signed checkpoints. A failing required gate denies authorization and records the reason.
One checkpoint for people, pipelines, and agents
Changes from people, pipelines, vendors, and AI agents can be routed through the same checkpoint and judged on the same terms.
Human accountability by design
Human-approval requirements are configurable by change type (Gate 7). AI analyzes; a named human validates; the system executes.
Who is accountable for a consequential change? A named human, set by policy per change type.
Recovery readiness is a condition of the decision
Backup restores data. Keystone is designed to help decide whether a change should run at all, and it treats recovery readiness as a condition of that decision.
- Typical today
- Separate backup process
- With a checkpoint
- Makes recovery readiness a precondition
The full method, including risk tiers, minimum decision evidence, metrics, and a 90-day rollout, is in the AI Change Governance Field Manual. Platform detail is on the Keystone page.
Related incident analysis: the Microsoft 365 outage of November 12, 2026 applies this governance model to a latent July update defect, representative-load validation and a proven recovery path under enterprise change volume.
AI change governance, answered
What is AI change governance?
AI change governance is the discipline of deciding, before execution, whether a routed change from a person, pipeline, or AI agent should run: tested on a safe group first, authorized by a named human where risk demands it, with a recovery point confirmed and a tamper-evident record kept. AuthorityGate Keystone is designed to deliver it as a checkpoint the customer owns.
What is Operational Resilience in the AI era?
Keeping critical services running through change at AI speed: governance decides who may change what, validation tests the effect in your environment, human experts decide what is new or uncertain, and recovery is ready first.
How is Keystone different from ITSM?
ITSM records intent and approval. Keystone is designed to add a tested outcome in your environment, a named approver where policy requires one, and recovery readiness before the change lands. It works alongside your ITSM process rather than replacing it.
Does Keystone replace backup?
No. Backup restores data. Keystone is designed to help decide whether a change should run at all, and it treats recovery readiness as a condition of that decision.
Who is accountable for a consequential change?
A named human, set by policy per change type. Gate 7 (SME Approval) is where configured approval requirements apply. AI analyzes; a named human validates; the system executes.
How do AI agents fit?
Changes proposed by AI agents can be routed through the same eight gates as changes from people and pipelines, so an agent's proposal is checked on the same terms before it is authorized.
What stage is the product?
Invitation-only early access through the Founding Members Program. Keystone is not generally available.
How can an enterprise help shape it?
Through the invitation-only Founding Members Program: members run Keystone in their own environments, work directly with the engineering team, and help shape features and the roadmap.
What framework mappings exist?
The AI Change Governance Field Manual maps the gates to NIST AI RMF, NIST CSF 2.0, ISO/IEC 42001, and the OWASP Top 10 for Agentic Applications. The mapping is labeled as AuthorityGate analysis, not an official crosswalk.
Who founded AuthorityGate?
Kevin Komlosy, Founder and CEO, who spent 30+ years validating change inside large regulated enterprises. AuthorityGate was founded on July 22, 2024, three days after the CrowdStrike outage.
Talk it through with the architect team
Keystone is in invitation-only early access through the Founding Members Program and is not generally available. If routed change in your environment needs a checkpoint, book 20 minutes.
Last updated .