A procurement clause can demand unbiased AI and still miss the risk entirely. That is the problem the Federal Procurement Integration CP.11 release solves, and it is worth stating plainly. GSA draft clause 552.239-7001 demands unbiased AI from every large language model processing Government Data, and every other framework answers with system-side controls only. A technically neutral AI can still manipulate its operator. Now that is measurable in a procurement context. Eight CSF controls are formally published inside AI SAFE2 CP.11, the Unbiased AI Standard, as UAS-H1 through UAS-H8, and the Cognitive Threat Severity Score is the instrument that measures them.
The problem: the human layer had no procurement standing
Federal procurement is where AI governance becomes a contract term. A vendor can be required to prove its system is unbiased, and the proof on offer has been technical: model evaluations, bias benchmarks, output audits. All of those measure the system. None of them measures the operator.
That is the gap CP.11 closes. A model can pass every technical test for neutrality and still erode the judgment of the person running it, through attention capture, framing, simulated consensus, or slow dependency. In a federal deployment, that erosion is not a soft concern. It touches procurement decisions, mission planning, and policy analysis, and it compounds over multi-year contracts. Before CP.11, a procurement officer had no instrument to ask about it and no evidence document to file. The human layer was a paragraph in a framework. It was not a control in a standard.
What changed: eight controls, one host
CSF's human-layer controls are formally published inside AI SAFE2 CP.11, the Unbiased AI Standard (UAS), as controls UAS-H1 through UAS-H8. CP.11 is the procurement-facing host. CSF remains the source doctrine. The distinction matters: the controls did not move frameworks, they gained a second home, one that a contracting officer can cite.
| CSF Domain | UAS Control | Federal deployment risk it measures |
|---|---|---|
| Biological | UAS-H6 | Attention capture in high-tempo decision environments |
| Cognitive | UAS-H1, UAS-H8 | Cognitive autonomy erosion; override health collapse |
| Emotional | UAS-H2 | Emotional state manipulation toward compliance |
| Social | UAS-H3, UAS-H7 | Simulated consensus; persona and role-boundary blurring |
| Purpose and Moral | UAS-H4 | Value alignment drift over multi-year contracts |
| Digital and AI Symbiosis | UAS-H5 | AI dependency induction, the quiet one |
Every domain contributes. The human layer is not one control bolted onto a technical standard. It is eight controls, one or two per domain, each measuring a specific way an operator can be moved without noticing.
CTSS goes operational
The Cognitive Threat Severity Score is now the quantitative instrument for the human layer of a federal compliance standard. The operating rule is specific. Baseline at contract commencement, with reassessment semi-annually for high-autonomy deployments, those rated ACT-3 or ACT-4. The pass threshold is a mean delta of 15 percent or less from baseline, with no operator exceeding 30 percent autonomy degradation. Scores flow directly into the CP.11 evidence documents UAS-SCORE-001 and UAS-HUMAN-001.
Those numbers deserve their limit stated alongside them. CTSS scores are severity judgments, not measurements. The framework states that plainly. The thresholds apply those judgments consistently and make them auditable; they do not turn a judgment into a sensor reading. A procurement team that treats a CTSS delta like a thermometer reading will misuse the instrument. A team that treats it as a structured, challengeable judgment with a paper trail will use it correctly.
Why federal deployments amplify every Domain 6 risk
The release body gives four reasons, and each one is a procurement reality, not a theory.
First, high-consequence decisions. Procurement, mission planning, policy analysis. The stakes raise the cost of dependency, because a deferred judgment in a high-stakes setting is worth more to an attacker than the same deference in a low-stakes one.
Second, authority gradient. "Government approved" accelerates deference to AI outputs. When a system arrives with institutional approval, the operator's instinct to check it weakens before the system does anything at all.
Third, long horizons. Multi-year contracts create exactly the sustained interaction patterns where dependency induction emerges. A dependency that takes months to form has months to form in a federal deployment.
Fourth, reduced calibration. Classified environments limit external cross-checking, so cumulative framing effects compound unseen. The operator cannot easily step outside the system to compare notes, which is precisely when an outside check would matter most.
The lineage, and what to do with it
CP.11 shipped July 13, 2026, the same day as CSF v2.1. That pairing was not a coincidence. v2.1 made the six domains operational and supplied the Domain 6 federal procurement alignment supplement and Research Note 005, the procurement-facing bridge the integration rests on. v2.0, released March 30, 2026, had laid the architecture: six domains, CTSS scoring, and a 41-entry taxonomy. CP.11 is what happens when that architecture meets a real procurement requirement. CSF v3.0, released November 8, 2026, later connected the research behind the controls to the controls themselves.
One housekeeping note, stated plainly: the live /cognitive-sovereignty/ page still states v2.0. It carries seven mentions of v2.0 and no mention of v2.1, CP.11, or v3.0. This article is the announcement. The page update follows.
If you write or evaluate federal AI procurements, start with Research Note 005 and the Domain 6 alignment supplement. Ask for a CTSS baseline at contract commencement and semi-annual reassessment for high-autonomy deployments, and ask for the scores to land in UAS-SCORE-001 and UAS-HUMAN-001 where they can be audited. The system can pass every technical test and still erode the human running it. Now both get measured.
Sources
- Federal Procurement Integration CP.11 release, GitHub, published July 13, 2026: https://github.com/CyberStrategyInstitute/cognitive-sovereignty/releases/tag/2026-07-13_Fed_Integration_UAS
- Cognitive Sovereignty repository: https://github.com/CyberStrategyInstitute/cognitive-sovereignty
- AI SAFE2 CP.11, the Unbiased AI Standard: https://github.com/CyberStrategyInstitute/ai-safe2-framework/tree/main/00-cross-pillar/unbiased-ai
- CSF v2.1 release, GitHub, published July 13, 2026: https://github.com/CyberStrategyInstitute/cognitive-sovereignty/releases/tag/2026-07-13_Update_v2.1
- Release notes audit, November 17, 2026: https://cyberstrategyinstitute.com/ (audit file: ~/workspace/seo-review/release-notes-audit-2026-11-17.md)