RAG Poisoning and EU AI Act Article 10: Data Governance Is Not Optional for Retrieval Pipelines reports that Article 10 compliance cuts RAG poisoning success from 95% to 20%, showing governance measurably reduces retrieval attacks. Outcome engineers must treat data governance as integral to retrieval pipelines—provenance, filtering, and audit controls change what an agent can safely retrieve and act on (Principles 02, 10, 14).

OpenAI adds plugin system to Codex to help enterprises govern AI coding agents unveils installable, versioned plugins for Codex so enterprises can control agent tool access and integrations via policy-controlled catalogs. This moves agent governance from prompt-era hacks to managed plugin registries and versioning—build registries, CI checks, and runtime policy enforcement around your agent toolchain (Principles 10, 11).

LangChain framework hit by several worrying security issues — ‘Each vulnerability exposes a different class of enterprise data’ details multiple LangChain vulnerabilities that leak enterprise data and risk downstream apps, requiring urgent patches and mitigations. If your stacks rely on LangChain or similar frameworks, add dependency audits, runtime defenses, and provenance tracing—framework bugs propagate through agent orchestration (Principles 10, 14).

Gitleaks creator returns with Betterleaks, an open source secrets scanner for the agentic era launches Betterleaks, an open-source secret scanner tuned to catch leaked credentials and hard-coded keys in AI-assisted development. Embed Betterleaks into CI, agent pipelines, and pre-commit hooks—secret scanning belongs in your Gate and Immune System to stop agents from leaking or committing secrets (Principles 15, 14).

Don’t YOLO Your File System introduces Jai, a one-command Linux sandbox that protects home directories with copy-on-write overlays to shrink agent blast radius. Use Jai or equivalent sandboxes to isolate agent processes during development and testing—sandboxing is a cheap, effective pattern for reducing accidental persistence and speeding safe iteration (Principles 07, 14).