What the FY2027 request buys
Verbatim from the R-2A exhibit for project MSL-05 of PE 0602025E. This is the budget justification's own description of work that has not happened yet — the one thing no other level of the budget carries.
- Develop formal approaches for software adaptation and system behavior prediction in the field. - Create physics-based models of perception for high reliability AI/ML-based sensing in autonomous systems. - Integrate distributed control, game-theoretic incentives, and strategic reasoning to enable robust, controllable, and adaptive distributed intelligence. - Develop techniques to detect, contain, and neutralize harmful behavior by AI agents. - Develop AI systems to translate between human language descriptions and provably correct software. - Develop multi-level security for foundation models. - Enable generative, adaptive engineering of complex systems that redesign themselves to maintain effectiveness.
The FY 2027 increase reflects studies and concepts initiations including, but not limited to, applied research in developing multi-level security for foundation models, techniques to neutralize harmful behavior by AI agents, and formal approaches for software adaptation and system behavior prediction.
Three years, and no five-year plan
An R-2A activity publishes the prior year, the current year and the budget year. The FYDP outyears exist at project and program-element level and are deliberately absent here rather than inferred. Estimate types are colored and never summed into one figure.
| Fiscal Year | Estimate Type | Amount ($M) |
|---|---|---|
| FY2027 | Request | 39.7 |
This activity is 29% of project MSL-05's FY2027 request and 2.4% of PE 0602025E's. In the request year the activities under a project sum to it exactly; in the current year they under-cover it in about 9% of cases, so an activity's delta can legitimately exceed its parent's and the two must not be compared row to row.
14 activities in project MSL-05
Every R-2A line of this project, largest FY2027 request first. Linked where the activity has enough of its own narrative to carry a page; the rest are shown in full on the project page.