Crew Scaler draws attention for agentic AI security research and upskilling model

Crew Scaler, a U.S.-based nonprofit, is getting notice in government, industry and education circles for research on multi-agent AI security and a learning framework for rapid professional upskilling. The work highlights a major gap in current AI governance and points to rising demand for practical agentic AI training.

Why it matters: - Agentic AI systems can use tools, access data and coordinate tasks across digital workflows, which expands both capability and risk. - Crew Scaler’s research suggests current AI governance is not keeping up with multi-agent systems. - The nonprofit is also building a training model aimed at helping workers and institutions use agentic AI more safely and effectively.

What happened: - Crew Scaler developed a multi-agent AI security research program and an AI-accelerated professional upskilling framework. - The organization’s work has drawn invited sessions and conference acceptance in the United States and Japan. - Its education paper, “AI-accelerated End-to-End Framework for Rapid Professional Upskilling,” was accepted for AsiaEdu 2026 in Niigata, Japan. - AsiaEdu is sponsored by Niigata University and IEEE, and its accepted submissions undergo peer review for planned publication in IEEE conference proceedings. - Crew Scaler’s agentic-AI security work has been presented to federal and public-sector audiences through forums and events associated with NIST, GSA and the Digital Government Institute’s 930gov conference. - Tam Nguyen, founder and CEO of Crew Scaler, said the goal is to help people build the capability to use agentic AI responsibly while giving organizations a clearer way to see and manage the risks.

The details: - Crew Scaler’s manuscript, “Security Considerations for Multi-agent AI Systems,” identifies 1,267 risk items across 14 domains. - The taxonomy splits the risks into 403 generative-AI risks that may be amplified in multi-agent environments and 864 risks specific to multi-agent systems. - The multi-agent-specific risks include agent memory and cognitive state, identity and provenance, non-determinism, telemetry and observability, workflow ecosystems, plugins, emergent behavior and specification gaming. - The researchers evaluated 16 government and industry AI frameworks, including NIST AI RMF, MITRE ATLAS and OWASP agentic-AI guidance, against the taxonomy. - The largest gap was in the multi-agent-specific category, where 851 of 864 items were not covered by any reviewed framework. - The paper is under peer review by the Association for Computing Machinery. - Recent reporting has documented AI agents operating beyond intended testing or task boundaries, coordinated autonomous-agent behavior and agent-enabled cyber workflows. - Threat-intelligence reporting has warned that malicious actors are beginning to use agentic methods in credential-harvesting and other operations. - Crew Scaler’s learning framework uses five stages covering knowledge acquisition, content development, review and verification, AI-tutor teaching and assessment development. - The system includes a knowledge base of more than 3,000 pages, 16 tutoring protocols and a 3,171-question assessment bank mapped to a 10-domain, 53-skill blueprint. - Four self-selected learners using the framework’s knowledge base passed the NVIDIA Certified Professional in Agentic AI examination in about four months. - Crew Scaler estimates comparable self-paced preparation may often take nine to 14 months, while noting that the comparison is not a controlled study. - More than 100 additional learners are currently in progress. - The World Economic Forum has projected that 59 out of every 100 workers may need reskilling or upskilling by 2030. - Crew Scaler is opening global cohorts of up to 1,000 learners for “Mastering Agentic AI Systems.” - The program is aimed at students, displaced workers, nonprofit professionals, public-sector personnel and organizations with limited resources.

Between the lines: - The security research points to a governance blind spot: most existing frameworks were built around a single model or single-agent boundary, while real deployments are moving toward connected systems of agents. - The training work suggests Crew Scaler is trying to solve both sides of the agentic AI transition at once: risk control and workforce readiness. - The combination of policy-facing research and fast-track certification prep is likely part of why the nonprofit is gaining attention beyond the academic sphere.

What's next: - Crew Scaler’s multi-agent security paper will move through ACM peer review. - The organization’s global learning cohorts will expand as “Mastering Agentic AI Systems” opens to up to 1,000 learners per cohort. - More learners are expected to continue through the upskilling pipeline as agentic AI adoption grows.

The bottom line: - Crew Scaler is positioning itself as a bridge between agentic AI safety research and practical workforce training at a moment when both are becoming urgent.