Lessons Learned from the Consortium: Tool Use in Agent Systems
A shared taxonomy for AI agent tools could anchor APS risk assessments and procurement conversations as agentic AI deployment grows across government.
Key points
- NIST and CAISI have developed two prototype taxonomies for classifying AI agent tool use, covering functionality and access constraints.
- Taxonomies offer APS governance teams a structured vocabulary for assessing and communicating AI agent capabilities and risks.
- Work is preliminary and community-facing; NIST is inviting public feedback rather than issuing a finalised standard.
Summary
Following a January 2025 workshop of approximately 140 experts under the AI Safety Institute Consortium (AISIC), NIST has published early-stage taxonomies for classifying tools used in AI agent systems. The first taxonomy organises tools by function - perception, reasoning, and action - while the second addresses constrained access patterns, distinguishing read-only from write permissions and trusted from untrusted environments. The work is explicitly preliminary; NIST is soliciting community feedback rather than issuing binding guidance. The taxonomies are intended to support transparency and communication across the AI supply chain, and complement existing resources such as NIST AI 600-1.
Implications for Australian agencies
- Monitor Agencies tracking agentic AI deployments may want to monitor this work as NIST iterates toward a more mature taxonomy or standard.
- Consider AI governance and risk teams could consider whether the functionality and access-pattern taxonomies are useful inputs to existing agent risk assessment or procurement frameworks.
Implications are AI-generated. Starting points, not advice.
"Lessons Learned from the Consortium: Tool Use in Agent Systems" Source: NIST – AI News (topic 2753736) Published: 5 August 2025 URL: https://www.nist.gov/news-events/news/2025/08/lessons-learned-consortium-tool-use-agent-systems Following a January 2025 workshop of approximately 140 experts under the AI Safety Institute Consortium (AISIC), NIST has published early-stage taxonomies for classifying tools used in AI agent systems. The first taxonomy organises tools by function - perception, reasoning, and action - while the second addresses constrained access patterns, distinguishing read-only from write permissions and trusted from untrusted environments. The work is explicitly preliminary; NIST is soliciting community feedback rather than issuing binding guidance. The taxonomies are intended to support transparency and communication across the AI supply chain, and complement existing resources such as NIST AI 600-1. Implications for Australian agencies: - [Monitor] Agencies tracking agentic AI deployments may want to monitor this work as NIST iterates toward a more mature taxonomy or standard. - [Consider] AI governance and risk teams could consider whether the functionality and access-pattern taxonomies are useful inputs to existing agent risk assessment or procurement frameworks. Retrieved from SIMS, 18 May 2026.