Agent Readiness Scoring: A Capability Taxonomy for Onboarding Enterprise AI Agents into Regulated Operations
DOI:
https://doi.org/10.63282/3050-9262.IJAIDSML-V6I2P128Keywords:
Agent Governance, Capability Taxonomy, Graduated Autonomy, Reversibility Risk, Enterprise AI Onboarding, Scope Drift Detection, Sequential TestingAbstract
As enterprises move from single-assistant deployments to agent fleets, the binding constraint shifts from model quality to onboarding governance: deciding which agent may act, on what surface, under what oversight. Existing work optimizes individual agent architecture and benchmark task success, and existing enterprise controls reduce to static role allowlists or committee review, neither of which expresses the property that actually determines exposure, namely whether an erroneous action can be undone. This paper presents AGORA (Agent Governance, Onboarding, and Readiness Assessment), a framework developed during evaluation and enablement of approximately 65 AI agents exposing more than 100 distinct capabilities inside a Fortune 50 telecommunications environment. AGORA decomposes every agent capability into a four-dimensional signature over action scope, data sensitivity, effect reversibility, and human checkpoint requirement, and organizes signatures as a partially ordered lattice. A monotone map from this lattice to a graduated autonomy ladder yields the Agent Readiness Score and the permitted deployment tier, so that autonomy is granted per capability rather than per agent. Promotion between tiers is adjudicated by a sequential probability ratio test over observed action outcomes, replacing calendar-based or approval-based gating with an evidence-based stopping rule. Fleet-level redundancy is detected through Jaccard similarity over capability signatures with community detection on the resulting graph. Silent scope expansion after model updates is detected by Kullback-Leibler divergence between declared and observed tool-invocation distributions. Against manual review, static role-based control, calendar-staged rollout, and an LLM-judge risk classifier, AGORA achieved 0.934 precision and 0.958 recall in restricting capabilities warranting restriction, reduced unauthorized action attempts by 91.3 percent, and detected post-update scope expansion with a median lead time of 3.1 days.
References
[1] Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2017). Concrete problems in AI safety. arXiv preprint arXiv:1606.06565.
[2] Beduschi, A. (2019). Digital identity: Contemporary challenges for data protection, privacy and non-discrimination rights. Big Data & Society, 6(2), 1–6.
[3] Brundage, M., Avin, S., Wang, J., Belfield, H., Krueger, G., Hadfield, G., Khlaaf, H., Yang, J., Toner, H., Fong, R., Maharaj, T., Koh, P. W., Hooker, S., Leung, J., Trask, A., Bluemke, E., Lebensold, J., O'Keefe, C., Koren, M., … Anderljung, M. (2020). Toward trustworthy AI development: Mechanisms for supporting verifiable claims. arXiv preprint arXiv:2004.07213.
[4] Cabrera, Á. A., Epperson, W., Hohman, F., Kahng, M., Morgenstern, J., & Chau, D. H. (2019). FairVis: Visual analytics for discovering intersectional bias in machine learning. 2019 IEEE Conference on Visual Analytics Science and Technology, 46–56.
[5] Sandeep Kamadi, " Risk Exception Management in Multi-Regulatory Environments: A Framework for Financial Services Utilizing Multi-Cloud Technologies" International Journal of Scientific Research in Computer Science, Engineering and Information Technology (IJSRCSEIT), ISSN: 2456-3307, Volume 7, Issue 5, pp.350-361, September-October-2021. Available at doi : https://doi.org/10.32628/CSEIT217560
[6] Cummings, M. L. (2017). Automation bias in intelligent time critical decision support systems. Decision Making in Aviation, 289–294.
[7] Sivaramakrishnan Narayanan (2022). Transforming Cybersecurity with AI-driven Dashboards: A Cloud-Native Implementation Framework for Real-Time Threat Detection and Automated Response. International Journal of Future Innovative Science and Technology (IJFIST) , Vol. 5 No. 5 (2022): International Journal of Future Innovative Science and Technology (IJFIST) , pp. 9207-9217. https://doi.org/10.15662/IJFIST.2022.0505004
[8] Dwork, C., & Minow, M. (2018). Distrust of artificial intelligence: Sources and responses from computer science and law. Daedalus, 151(2), 309–321.
[9] Endsley, M. R. (2017). From here to autonomy: Lessons learned from human-automation research. Human Factors, 59(1), 5–27.
[10] Hind, M., Mehta, S., Mojsilovic, A., Nair, R., Ramamurthy, K. N., Olteanu, A., & Varshney, K. R. (2018). Increasing trust in AI services through supplier's declarations of conformity. arXiv preprint arXiv:1808.07261.
[11] Meesala, A. (2022). Adaptive Spread Anomaly Intelligence Framework (ASAIF): A cloud-native AI framework for real-time bid-ask spread anomaly detection and cross-venue liquidity risk intelligence. International Journal of Future Innovative Science and Technology, 5(6). https://doi.org/10.15662/IJFIST.2022.0506007
[12] Kim, B., Park, J., & Suh, J. (2020). Transparency and accountability in AI decision support: Explaining and visualizing convolutional neural networks for text information. Decision Support Systems, 134, 113302.
[13] Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model cards for model reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229.
[14] Lakshmi Kiran Meesala "AI-Driven Cyber Defense: A Hybrid Deep Learning Framework for Real-Time Threat Prediction and Response" International Journal of Scientific Research in Science, Engineering and Technology (IJSRSET), Print ISSN: 2395-1990, Online ISSN: 2394-4099, Volume 10, Issue 2, pp.916-930, March-April-2023. Available at doi: https://doi.org/10.32628/IJSRSET235845
[15] Raji, I. D., Smart, A., White, R. N., Mitchell, M., Gebru, T., Hutchinson, B., Smith-Loud, J., Theron, D., & Barnes, P. (2020). Closing the AI accountability gap: Defining an end-to-end framework for internal algorithmic auditing. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 33–44.
[16] Arun Meesala , " A Distributed Kafka-Centric Framework for High-Throughput Mid-Price Computation and Intelligent Time-Series Persistence in Financial Clouds" International Journal of Scientific Research in Computer Science, Engineering and Information Technology(IJSRCSEIT), ISSN : 2456-3307, Volume 9, Issue 2, pp.998-1006, March-April-2023. Available at doi: https://doi.org/10.32628/CSEIT2342442
[17] Rose, S., Borchert, O., Mitchell, S., & Connelly, S. (2020). Zero trust architecture (NIST Special Publication 800-207). National Institute of Standards and Technology.
[18] Lakshmi Kiran Meesala, " Modern Security Information and Event Management: Architecture, Analytics Pipelines, and Empirical Evaluation of SOC-Scale Threat Detection" International Journal of Scientific Research in Computer Science, Engineering and Information Technology (IJSRCSEIT), ISSN: 2456-3307, Volume 9, Issue 4, pp.995-1012, July-August-2023. Available at doi: https://doi.org/10.32628/CSEIT23564538
[19] Sculley, D., Holt, G., Golovin, D., Davydov, E., Phillips, T., Ebner, D., Chaudhary, V., Young, M., Crespo, J.-F., & Dennison, D. (2017). Hidden technical debt in machine learning systems. Advances in Neural Information Processing Systems, 28, 2503–2511.
[20] Sivaramakrishnan Narayanan (2023). Operationalizing Artificial Intelligence Security in the Cloud: A Practical Integration framework for Enterprise Risk Management. International Journal of Future Innovative Science and Technology (IJFIST), Vol. 6 No. 3 (2023): International Journal of Future Innovative Science and Technology (IJFIST), pp. 10611-10619. https://doi.org/10.15662/IJFIST.2023.0603002
[21] Shneiderman, B. (2020). Human-centered artificial intelligence: Reliable, safe and trustworthy. International Journal of Human–Computer Interaction, 36(6), 495–504.
[22] Zhang, J. M., Harman, M., Ma, L., & Liu, Y. (2020). Machine learning testing: Survey, landscapes and horizons. IEEE Transactions on Software Engineering, 48(1), 1–36.
