Netherite

Security work is moving from reactive incident response to proactive threat modeling, and agencies that build this into retainers gain a compounding trust advantage. The liability risk of promising absolute security is real, but the bigger risk is staying silent while clients face evolving attack surfaces and new compliance mandates. Bundling threat modeling with incident response creates a defensible service line that clients cannot easily replicate in-house.

Every security retainer carries an implicit liability ceiling: the gap between what an agency promises and what an attack surface can actually guarantee. Agencies that sell "we will keep you secure" absorb unlimited downside; agencies that sell defined detection, response, and remediation scopes cap their exposure while still charging recurring fees. The framework asks three questions before signing: what specific asset is protected, what detection window is promised, and who owns the residual risk when a novel attack path emerges. Cogent's VR-1 model maps attack paths across enterprise infrastructure, which reframes the deliverable from "prevention" to "path visibility," a bounded promise. Sentrint grades repository security and generates fix prompts, giving clients a measurable artifact rather than an assurance. Vaultak monitors and rolls back AI agent actions in production, another bounded scope. California SB 813 and AB 1405, signed September 9, 2026, formalize third-party AI audit expectations, which pushes agencies toward documented, auditable scopes instead of blanket guarantees.

Blast Radius Budgeting treats security scope as a function of how much damage a single compromised asset can cause, not how many assets exist. An agency protecting a 40-person client with one shared drive has a smaller blast radius than a 12-person client whose AI agents hold production database credentials. The framework asks three questions per engagement: what can be reached from the weakest credential, how fast can it be revoked, and who eats the loss if it is not. That third question is the pricing lever. Runtime governance layers such as Vaultak intercept agent actions and roll them back automatically, which shrinks the radius and justifies a lower liability premium; frontier reasoning models like Cogent map attack paths across the same infrastructure, which expands the billable discovery phase. California SB 813 and AB 1405, signed September 9, 2026, now formalize third-party audit expectations, so documented radius estimates become client-facing evidence rather than internal notes.

The Trust Premium Stack treats security capability as a pricing lever rather than a cost center. Each layer an agency can credibly demonstrate (device policy enforcement, code scanning, runtime agent governance, incident response) raises the ceiling on what a client will pay for the same deliverable, because the buyer is pricing risk transfer, not hours. The stack only works when every layer is evidenced: a claim without a scan report or a policy dashboard is a discount waiting to happen. Consider the governance gap now opening around AI agents. Vaultak intercepts and can roll back agent actions in production, which gives an agency a concrete artifact to show a client whose automation touches customer data. Pair that with a repository scanner such as Sentrint producing a graded report, and the retainer conversation shifts from rate card to risk coverage. The ceiling is real but finite: no layer justifies promising absolute security.

Scope every security engagement as detection, evidence, and response support, and never contract for absolute protection.

Buy the enforcement layer before the detection layer whenever an agent holds write access to a client system.