Context Hub vulnerable to supply chain attacks, says tester. Unvetted Context Hub docs can hide poisoned dependencies that coding agents silently inject, revealing a dangerous supply-chain attack vector. Outcome engineers must treat external context artifacts as untrusted inputs — add provenance checks, dependency vetting, and sandboxed execution to your intake pipeline (Principles 02, 14, 15).

OpenAI adds plugin system to Codex to help enterprises govern AI coding agents. OpenAI ships installable, versioned plugins for Codex so teams can control tool access, integrations, and policies via a managed catalog. This shifts enforcement from prompts to gateable artifacts — treat plugin catalogs as the new policy surface for access control, auditing, and least-privilege (Principles 10, 11).

RAG Poisoning and EU AI Act Article 10: Data Governance Is Not Optional for Retrieval Pipelines. Analysis shows Article 10 compliance reduces RAG poisoning success dramatically, surfacing that provenance, labeling, and pipeline controls materially lower attack surface. Outcome engineers should bake dataset governance, provenance telemetry, and automated poisoning-detection into RAG systems to protect Ground Truth and meet regulation (Principles 02, 10, 14).

Agent-to-agent pair programming. Loop runs two coding agents side-by-side in tmux so agents can pair, review, and steer each other’s output for faster, higher-quality code. Adopt harnesses like this to operationalize multi-agent review workflows, reduce single‑agent failure modes, and make team norms explicit in agent interactions (Principles 03, 09).

LangChain framework hit by several worrying security issues — ‘Each vulnerability exposes a different class of enterprise data’. Multiple LangChain vulnerabilities surface classes of enterprise data leakage and require immediate patches and mitigations. If LangChain is in your stack, treat framework updates as security-critical, add runtime guards and secrets scanning, and enforce an immune-system posture for frameworks that sit inside your agentic workflows (Principles 14, 15).