Curated from knowledge graph (201 nodes, 244 edges) · All articles published within the last 7 days

• Canvas owner confirms cybersecurity incident — Instructure, the company behind Canvas LMS used by thousands of higher education institutions, confirmed a data breach affecting user names, messages, email addresses, and student ID numbers. Critical infrastructure reminder for campus IT leaders. 🔗 Graph: uc-san-diego, canvas-lms, ai-security 📅 Published: 2026-05-08 📰 https://www.highereddive.com/news/instructure-confirms-cybersecurity-incident/819586/

• Authorization Propagation in Multi-Agent AI Systems: Identity Governance as Infrastructure — New arXiv research argues that multi-agent systems create a distinct authorization problem beyond prompt injection: maintaining authorization invariants as non-human principals retrieve data, delegate tasks, and synthesize results across changing boundaries. Critical reading for enterprise AI deployments. 🔗 Graph: agentic-ai, ai-governance, litellm-enterprise 📅 Published: 2026-05-08 📰 https://arxiv.org/abs/2605.05440

• AgenticRAG: Agentic Retrieval for Enterprise Knowledge Bases — Researchers propose a practical agentic harness for retrieval and analysis over enterprise knowledge bases, layering lightweight reasoning on top of existing enterprise search infrastructure. Reduces overdependence on fixed candidate sets chosen deep in the retrieval process. 🔗 Graph: tritongpt, onyx, enterprise-data-agent 📅 Published: 2026-05-08 📰 https://arxiv.org/abs/2605.05538

• Anthropic updates Claude Managed Agents with three new features — Anthropic launched Claude Managed Agents simplification tools last month, and this update adds three new capabilities including streamlined deployment workflows and expanded integrations. Direct relevance to TritonAI's enterprise Claude usage. 🔗 Graph: anthropic, claude, agentic-ai 📅 Published: 2026-05-07 📰 https://9to5mac.com/2026/05/07/anthropic-updates-claude-managed-agents-with-three-new-features/

• LaTA: A Drop-in, FERPA-Compliant Local-LLM Autograder for Upper-Division STEM Coursework — New open-source autograder runs entirely on commodity on-premises hardware, assuming LaTeX document input. Addresses the privacy-cost-capability tension in educational AI: cloud models expose sensitive student work, while local models preserve privacy but remain less reliable. Directly relevant to UC's 73K+ user deployment model. 🔗 Graph: tritonai, higher-ed-ai, ai-governance 📅 Published: 2026-05-08 📰 https://arxiv.org/abs/2605.05410

• Knowledge Graphs, the Missing Link in Agentic AI-based Formal Verification — Research exploring how LLM-generated SystemVerilog Assertions can benefit from knowledge graphs for handling ambiguous specifications. Positioning KG-augmented reasoning as essential for reliable hardware verification workflows. 🔗 Graph: knowledge-graphs, agentic-ai, model-context-protocol 📅 Published: 2026-05-08 📰 https://arxiv.org/abs/2605.06434

• Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation — Safety research identifying a critical blind spot: harmful content may appear in chain-of-thought reasoning traces even when final answers appear safe. Tests both reasoning and answer stages under a unified twenty-principle safety rubric. 🔗 Graph: ai-security, ai-governance, claude 📅 Published: 2026-05-08 📰 https://arxiv.org/abs/2605.05678

💡 Signal: This week's research output is heavily weighted toward agentic AI safety and governance—reflecting the field's maturation from "can we build it" to "can we deploy it responsibly." The cluster of papers on authorization propagation, safety failures in reasoning traces, and knowledge-graph-augmented verification suggests the research community is grappling with the same operationalization challenges Brett's team faces at scale.