Insights

When capture wants AI and security wants evidence

Align AI discriminators to controls you can show — approved tools, review rules, pipeline policy — so proposal language and delivery practice match.

Capture wants an AI discriminator. Security wants residual risk stated clearly. Both can be right — and both can lose if the volume says one thing and the pipeline does another.

We see the pattern on pursuits where “AI-enabled DevSecOps” lands in a win theme before anyone has listed approved tools, written a data boundary, or confirmed that AI-assisted commits still pass the same scanners as everything else. The color team likes the modernity. The ISSO-adjacent reviewer asks for evidence. Orals become a negotiation between marketing fluency and engineering fact. Delivery inherits the overpromise.

This sits next to a broader capture problem we write about elsewhere: pursuit week should be execution, not archaeology. An AI claim with no evidence pack is another form of unreadiness — just louder.

Align the claim to the control narrative

A discriminator is a proposal claim meant to differentiate. It only helps if it is defensible. In this domain, defensibility usually means: you can show how AI sits as a system of suggestion, while humans and the pipeline remain the system of record.

The control narrative is how you explain — in proposals, System Security Plans where relevant, and program reviews — which controls exist, how AI tools are constrained, and what residual risk decision makers are accepting. It is not a slogan about innovation. It is a coherent story technical staff can defend without inventing past performance or implying certifications the company does not hold.

When capture and security are misaligned, two failure modes dominate:

  1. Proposal overreach — Capabilities the pipeline cannot demonstrate; buzzwords without an evidence pack.
  2. Security as a late gate — AI accelerated volume of change; findings show up in pre-prod while the volume already claimed maturity.

Neither is fixed by another marketing slide. Both are fixed by putting engineering practice and proposal language in the same workshop — then writing only what you can show.

Practical moves

These are checks we use when the pursuit calendar wants an AI story and delivery is still catching up. They are not a universal scoring formula, and they are not legal or procurement advice.

1. Inventory before you draft the discriminator.
List AI touchpoints in delivery and in the proposal factory: IDE assistants, chat tools, test generators, RFP analyzers, ops copilots. Note owner, data destination, and approval status. If capture cannot name the tools delivery actually uses, the discriminator is fiction.

2. Map each AI claim to a control you can exhibit.
For every sentence that mentions AI-assisted delivery, ask: Where is the approved tool list? Where is the data-handling or prompt-hygiene one-pager? Where do pipeline policies, scanner configs, or review rules for high-risk paths live? If the answer is “we’ll build that after award,” cut or soften the claim now. Evaluators and orals panels notice the gap even when the matrix is green.

3. Prefer evidence-backed differentiators over logo tours.
Stronger discriminators sound like operating practice: mandatory review on auth and infrastructure-as-code paths; secret scanning on by default; enterprise AI seats with contractual data-use terms; subcontractors under the same AI rules. Weaker ones sound like product brochures. We do not invent client names, win rates, or Authority to Operate outcomes to inflate either. Method and artifacts carry the technical approach when past performance belongs elsewhere.

4. Put residual risk in plain language.
Security partners are not trying to kill the chase when they ask what the model can still get wrong: leaked context, vulnerable patterns, license contamination, over-trust. State what you constrain, what you monitor, and what you accept. Vague assurance reads as evasion. Clear residual risk reads as adult program management.

5. Reuse the same diagrams in volumes and security appendices.
Show where AI suggests and where authority sits (human review + pipeline gates). Capture-ready documentation that both proposal and security teams recognize reduces last-week archaeology. Update boilerplate when practice changes — marketing, delivery, and security should say the same thing.

A ninety-day sketch many teams can adapt: days 1–30 see clearly (inventory, boundary draft, top high-risk workflows); days 31–60 wire the gates (pipeline confirmation, review rules, hygiene training); days 61–90 align the story (SOPs and proposal language). Adjust to your calendar; do not treat the sketch as a guaranteed outcome.

Who this is for

This is for federal contractors whose capture teams feel pressure to modernize the narrative and whose security partners need that narrative to stay true. It is educational. Mentions of FedRAMP or Authority to Operate here are conceptual — we do not claim Jiara performs ATO or holds FedRAMP authorization. Engagement-type work we typically help with looks like proposal control-narrative alignment, AI tooling policy paired to pipelines, and documentation packages that keep suggestion vs. authority visible — the same capture-ready documentation discipline that keeps pursuit week from becoming archaeology.

Read the companion primer DevSecOps in the AI World for the seven-move framework and starter checklist. If you want a readiness conversation about aligning discriminators to controls before the next color team, use Contact.