Where should AI agents stop and ask for human approval? A runtime risk framework
Invisible_act1988 · reddit · 2026-09-29
A developer working on AI agent security argues the hard problem isn't tool access but runtime permissions: what is an agent actually allowed to do?
- Low-risk actions (reading invoices, creating drafts, updating customer info) can run autonomously
- High-risk actions (sending money, changing bank details, exporting customer data, deleting records) need approval
- Riskiest of all are action chains that are safe individually but dangerous in sequence
The author argues the answer isn't simply fewer permissions, but evaluating action + context + risk + what the agent has already done, pausing for human approval when needed.
Related event: Runtime authorization emerges as key challenge for AI agent safety(3 posts)→
More from coding & agent
- Generic API-catalog MCP server vs one tool per job: a context-bloat experiment — MandeepKaur9407 · 2026-11-16
- Berkeley paper: LLMs answer with stale values even when the update is in context — ombre-ai · 2026-11-16
- Effective HTML: 4.2k-star agent skills turn wireframes into self-contained interactive prototypes — tom_doerr · 2026-11-16
- skills-gate: one gate skill fronts 148 agent skills at just 389 tokens — Distinct-Pie2389 · 2026-11-16
- Open-weight Karpo-Decision-31B tops Decision Index at 73.9, one forward pass per answer — CodeByPoonam · 2026-11-16
- shadcn boosts @uisdk: one-shot UI rebuild that's instantly E2E-testable by AI — shadcn · 2026-11-16