Post by Afterburner Wind Tunnel Services, LLC

Your Weapons Program Has AI In It. Your Safety Process Doesn't Know That Yet. Your weapons program probably has AI in it. Your SSPP almost certainly doesn't reflect that yet. it's a structural problem. The traditional MIL-STD-882E hazard analysis workflow was built for deterministic software. You define the behavior, you test the behavior, you document the residual risk. That process works. But ML models don't behave deterministically, and the safety frameworks most programs are running today weren't designed with that in mind. The DoD CTO acknowledged this directly. In January 2026, they published the Machine Learning System Safety Engineering Guide (MLSSEG), built on MIL-STD-882E w/CHANGE 1, specifically because AI/ML integration into defense systems has outpaced existing safety practice. One of the more significant callouts in that guide: ML systems cannot achieve Software Criticality Index Level 1 assurance. There is no SFCI 1 column in the Appendix A LOR Matrix for ML — because probabilistic models cannot reach that assurance tier by design. If your system has a critical function, it needs a non-ML interlock or watchdog. The guide also introduces a task most legacy SSPPs don't have: Data Hazard Analysis. DHA requires you to evaluate data provenance, label quality, class imbalance, and data poisoning risk. None of that exists in a standard PHA or SHA template. And then there's model drift — the reality that ML performance degrades as operational environments diverge from training data. That's an ongoing safety liability with no equivalent in traditional software safety practice. Meanwhile, the DoD FY25-26 Software Modernization Implementation Plan is accelerating AI fielding timelines across the department. Programs are moving fast. The safety frameworks are catching up. That gap is a live program risk — both from a genuine field safety standpoint and from an audit/compliance standpoint, as DAAs and program offices are increasingly asking about AI-specific hazard analysis. If you're on an AI-enabled program and your SSPP hasn't been updated to address MLSSEG requirements, that's worth a hard look before your next milestone review. Happy to talk through what that gap assessment looks like in practice.