INVARIANCE · Independent execution authorization for AI agents
AI agents now move money, change supplier details, issue prescriptions and grant access. They hold the credentials to do it. What they do not hold is independent authorization.
The launch partner program extends to healthcare and veterinary AI agent providers.
INVARIANCE opens its launch partner program for finance and insurance AI agent providers.
The problem
An agent with API credentials and delegated permissions can initiate a payment, change a beneficiary or grant privileged access. Access controls establish what an agent can reach. They do not establish whether a specific action, at a specific moment, is authorized to execute.
When the agent is also the system deciding whether its own action is safe, there is no independent check at all. The authority sits with the actor.
The system taking the action cannot be the system authorizing the action.
Illustrative scenario · financial operations
An agent receives an instruction to update a supplier's bank details. Minutes later, the same workflow attempts a $48,000 payment to the new account. The agent holds valid credentials for both actions.
At the execution boundary, INVARIANCE finds no independent confirmation of the beneficiary change. The payment is denied before it takes effect, and the denial is recorded as evidence.
The agent had the credentials. It did not have independent authorization. The $48,000 never left.
Candidate action
independent confirmation: absent
deny · policy FIN-07 · gate closed · denial recorded
How INVARIANCE works
INVARIANCE provides an independent execution boundary between what an agent decides and what it executes. Each consequential action is authorized or denied before it takes effect.
The boundary sits outside the agent and outside the system that proposed the action. The agent cannot authorize itself.
Authorization applies to one action, with its amount, counterparty and context, at the moment of execution. It is not a standing permission.
If the check cannot run, the action does not. Silence, timeout and error all resolve to no execution.
Every decision, permit and denial alike, leaves a record that can be verified independently after the fact.
Where it applies
Payments, beneficiary changes and approvals initiated by agents in finance workflows.
Underwriting decisions taken by agents within delegated authority and policy limits.
Access grants and privileged changes requested by agents in operational systems.
Treatment, prescribing and referral actions taken by clinical AI agents. VETPASS, a live clinical AI platform, is the original proving ground for INVARIANCE.
For AI agent providers
Finance, insurance, healthcare and veterinary agent providers ship INVARIANCE with their agents. Their customers get agents that cannot act beyond their authority, with tamper-evident evidence of every consequential decision.
INVARIANCE is independent execution authorization infrastructure for consequential AI: the control plane for safe AI action. Governance establishes the rules. INVARIANCE independently authorizes each agent action against them before it executes. Where INVARIANCE sits
Launch partners
INVARIANCE is selecting a small number of launch partners among finance, insurance, healthcare and veterinary AI agent providers.
Enterprises deploying agents and infrastructure partners are welcome too, through the contact page.