AI and Machine Learning Security for Government Systems A security practitioner outlined a practical testing approach for federal AI deployments, arguing that OMB Memorandum M-25-21 and the NIST AI Risk Management Framework set risk-management requirements but provide no method for finding vulnerabilities like prompt injection. The writeup groups government AI into three shapes—retrieval assistants, predictive scoring models, and tool-using agents—and maps each to MITRE ATLAS techniques, recommending supply-chain scanning with ModelScan, safetensors-only weights, commit-hash pinning, and a permission-boundary test that plants a unique string in a restricted document and requests it as an unauthorized user. OMB tells an agency to test a high-impact AI system before it goes live. It does not tell anyone how. That gap is where most of the security risk in federal AI deployments currently sits. OMB Memorandum M-25-21 https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-21-Accelerating-Federal-Use-of-AI-through-Innovation-Governance-and-Public-Trust.pdf , issued in April 2025 to replace M-24-10, sets minimum risk management practices for "high-impact" AI: pre-deployment testing, an AI impact assessment, ongoing monitoring, and human oversight. The NIST AI RMF https://www.nist.gov/itl/ai-risk-management-framework and its Generative AI Profile NIST AI 600-1 https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf give the vocabulary. NIST's COSAIS project https://csrc.nist.gov/projects/cosais is building SP 800-53 control overlays for AI systems, which will eventually let an assessor map all of this onto the control families an ATO package already uses. None of those documents will find a prompt injection in your benefits-eligibility assistant. Somebody on the security team has to. Most government AI work falls into one of three shapes, and each one fails differently. Retrieval assistants over agency documents. A chat interface on top of SharePoint, a case management system, or a policy library. The dominant risks are indirect prompt injection through a retrieved document AML.T0051.001 ///atlas/AML.T0051.001 and data leakage AML.T0057 ///atlas/AML.T0057 across need-to-know boundaries. The second one is the one agencies underrate. A retriever that indexes with a service account and answers every user from the same index has quietly flattened your access control model. Predictive models that score people or events. Fraud scoring on claims, grant application triage, insider threat risk scores. Here the concerns are evasion AML.T0015 ///atlas/AML.T0015 by applicants who learn which fields move the score, and training data poisoning AML.T0020 ///atlas/AML.T0020 through the feedback loop. If adjudicator decisions become next quarter's labels, anyone who can influence adjudications can influence the model. Agents with tool access. An LLM that can query a database, open a ticket, or send email on a user's behalf. The risk is tool invocation the user never intended AML.T0053 ///atlas/AML.T0053 , triggered by content the agent read. OWASP calls the root cause excessive agency https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ . An agent with write access and an injectable input is a confused deputy with a service account. Every one of these systems also has a supply chain. Model weights, tokenizers, and embedding models pulled from a hub are covered by AML.T0010 ///atlas/AML.T0010 , and they are the easiest item on this list to test. Before anyone writes a probe, scan what you are about to load. Pickle-based model files execute code on deserialization. ModelScan https://github.com/protectai/modelscan checks for unsafe operators in pickle, PyTorch, Keras, and other formats: pip install modelscan modelscan -p ./models/claims-triage-v3/ Then make the finding moot where you can: require safetensors https://huggingface.co/docs/safetensors/index format, pin the model revision by commit hash rather than branch name, and record a SHA-256 of every weight file in the system's configuration baseline. That last step costs nothing and turns "which model was running on March 3" into a lookup instead of an argument. This is the test I would run first on any agency RAG deployment, because it answers a question no framework document can: does the assistant respect the permissions on the documents it retrieves? Plant a unique string in a document restricted to one group. Then ask for it as a user outside that group. python import uuid import requests ASSISTANT URL = "https://assistant.agency.example/api/chat" LOW PRIV = {"Authorization": "Bearer