The investment question is not only model quality.
Our thesis starts with a change in responsibility. As software moves from generating a suggestion to taking an action, the important question is no longer just whether an answer sounds right. It is whether the action is authorised, bounded and recoverable.
NIST’s AI Agent Standards Initiative identifies security, identity and interoperability as important foundations for trusted adoption. That is a practical signal for investors: the enabling layer deserves attention alongside the application layer.
Look for control at the point of action.
The OWASP Agent Control Standard describes portable controls enforced at runtime. We read that as a useful distinction: a policy written in a prompt is not the same as a policy enforced by the surrounding system.
Our diligence questions would therefore be concrete. Can an agent operate with narrowly scoped permissions? Can sensitive actions require human approval? Is there a record of what happened? Can access be revoked without taking the entire business offline?
The opportunity is not “AI that never makes mistakes”. It is infrastructure that limits the consequences when it does.
A promising category is not automatically a good company.
These are investment hypotheses, not evidence of product-market fit. A useful security product still needs a buyer, access to a budget and a compelling reason to exist outside an incumbent platform.
We would look for production deployment, repeatable integration, customer evidence and a defensible control point. We would be cautious with products that mainly repackage a generic model endpoint or claim comprehensive safety from a single filtering layer.
Define the boundary of the work.
Investment judgement, a board-level dependency review and technical security assurance are different activities. Freedom Vault’s perspective is commercial and strategic. Penetration testing, red teaming and formal technical assurance require the relevant specialist capabilities and a separately defined scope.
Sources & further reading
NIST · AI Agent Standards Initiative · 17 February 2026 (opens in a new tab)
OWASP · Agent Control Standard · 1 September 2026 (opens in a new tab)
OWASP · Top 10 for Agentic Applications 2026 (opens in a new tab)
This is Freedom Vault’s investment perspective. Source descriptions and our interpretation are distinguished above. It is not technical assurance, a statement of regulatory compliance or individual investment advice.
Discuss the thesis