Every agent passes its own governance review — but who checks agents against each other?
HimalAIan · reddit · 2026-09-30
A practitioner argues AI governance is stuck on a per-agent view — access, approvals, logging — while agent-to-agent interactions go unchecked. Example: a procurement agent allowed to expedite orders books freight against a cost center a finance agent froze an hour earlier; both followed their rules, but the company made a commitment it didn't want. Config fixes work for deterministic workflows, but agents decide at runtime across team-owned tools, and change management reviews each one separately. The author poses three open questions: do reviews cover inter-agent interactions, who decides conflicts between approved agents, and is anyone testing for this before incidents?
More from coding & agent
- Microsoft launches Azure canvases: manage Azure resources inside GitHub Copilot — DanWahlin · 2026-09-30
- GPT-6.1 Sol fixes 44 of 105 planted bugs for $6.56, matching Astra at a fraction of the cost — PawelHuryn · 2026-09-30
- How Anthropic made claude.ai 3x faster in two weeks with Claude, merging 3,000+ changes — trq212 · 2026-09-30
- Anthropic engineer's prompt trick: tell Claude "be braver" to unlock bolder behavior — trq212 · 2026-09-30
- Agent Calls You, Asks Three Questions, Answers Retrieved via API — gregmushen · 2026-09-30
- Qwen Code v0.24.7 nightly adds local workspace-agent collaboration — qwen-code-ci-bot · 2026-09-30