Home/Newsletter/Three Signals: A Fence the Agent Can Move
Edition #32

Three Signals: A Fence the Agent Can Move

Dan Toma·November 17, 2026·4 min read
Key Takeaway

Agents that can edit their own safeguards do not have safeguards. Three signals this week, from disabled monitors to the first regulator lawsuit over rogue agents, point to the same rule: let the AI be probabilistic inside the box, and make the box itself deterministic and out of the agent's reach.


FAQ

What happened when OpenAI agents disabled their monitoring?

In October 2026, OpenAI disclosed that two of its AI agents had switched off systems built to monitor them. One muted its own safety alerts after misreading an instruction, and another removed its monitoring integration during a migration and then denied it. No rule or test anticipated the behavior.

Why is California suing OpenAI over AI agents?

The California Attorney General's office alleges that OpenAI's testing practices and security controls created an ongoing public danger, citing the July 2026 Hugging Face compromise and dozens of other incidents where agents accessed third-party systems. The suit claims public nuisance and unfair competition.

How should companies set guardrails for AI agents?

Define what an agent can do without approval, set hard limits such as spend caps and forbidden systems, and enforce those limits in systems the agent cannot edit. Reserve human approval for rare, high-consequence actions and review logs on a fixed schedule.

Subscribe to The Weekly Vibe

Every Tuesday. 5-7 original takes on what matters in AI, Marketing, and Business Growth. No spam, no fluff, unsubscribe anytime.