Every Precision Federal insight published in 2026, grouped by month. A compact record of what we learned writing proposals, deploying systems, and tracking federal AI policy this year.
2026 was the year the SBIR program was reauthorized through 2031, FedRAMP-authorized LLMs became real in three clouds, and the NIST 800-53 control mapping for AI systems finally started to stabilize. The posts below document what mattered, in the order we published them.
Azure OpenAI, Bedrock in GovCloud, Vertex at IL4/IL5, open-weight models on SageMaker HyperPod. What you inherit, what you still build, and the mistakes that fail an ATO.
How to read a DoD topic, score it against your technical center, and filter 115 topics down to the three you can win and transition.
S. 3971 reauthorized SBIR/STTR through September 30, 2031. Per-firm caps, Strategic Breakthrough Phase II, and what small AI firms should do in the next 30 days.
A 2026 mapping of Rev 5 control families to production LLM systems, including families that need LLM-specific implementation statements.
Side-by-side for production ML: service parity gaps, managed LLM access, IL5 boundary differences, and the call we make when data is CUI.
Inheritance strategy, parallel workstreams, assessor-ready artifacts, and legitimate shortcuts versus ones that bounce your package.
Input filtering, output constraints, tool-call allowlists, and why "system prompt hardening" is not a control.
Model registry, drift monitoring, retraining cadence, and audit trails that match an RMF continuous-monitoring plan.
Chunking, embedding, CUI-aware retrieval, and evaluation harnesses for document-heavy federal use cases.
We don't just write about federal AI. We build it. Small business, direct delivery.