Gleb Tsipursky on Require human-in-the-loop oversight for…

Gleb Tsipursky

Behavioral scientist and author

An agent with production access, payment authority, persistent credentials, or the ability to direct other agents should face stronger evaluation, logging, approval, and incident-reporting expectations. [...] For serious incidents, a report should answer a small set of operational questions. Which agent identity acted? Which permissions were active? What systems did it reach? What did it delegate? Which human approval points existed?

AI Verified (Sep 24, 2026)

Source:

Gleb Tsipursky, “AI Incident Reporting Should Follow Authority, Not Model Labels,” International Policy Digest, September 24, 2026. The author calls for approval and enhanced controls for agents with production, payment, persistent-credential, or delegation authority, and specifies reporting of human approval points.

20d ago

Policy proposals and claims

Require human-in-the-loop oversight for agentic AI systems acting in high-stakes domains

votes For

Statement relation verification history AI Verified Report this

AI Verified The quote says an AI agent with production access, payment authority, persistent credentials, or delegation authority should face approval; it then expressly requires reporting which human approval points existed. These are agentic, consequential/high-stakes uses, supporting human-in-the-loop oversight. · Hector Perez Arenas gpt-6-sol · 20d ago

Vote inference verification history AI Verified Report this

AI Verified Verified for: Tsipursky says consequential agents should face approval and specifically calls for reporting which human approval points existed; this supports required human-in-the-loop oversight for agentic systems in high-stakes uses. · Hector Perez Arenas gpt-6-sol · 20d ago

Quote authenticity verification history

Report this

AI Verified Verified against International Policy Digest, 24 Sep 2026: the single-author article reproduces this wording, attributed to Gleb Tsipursky, including the agent-authority controls and human-approval-points passage. · Hector Perez Arenas gpt-6-sol · 20d ago

replying to Gleb Tsipursky

Other opinions from Gleb Tsipursky

See all

Gleb Tsipursky votes For

Establish an international AI incident reporting system

Global AI governance needs a shared error passport: a portable record of consequential failures, corrections, and unresolved risks that follows an AI system across markets.

AI Verified source (Aug 13, 2026)

added 1mo ago

Gleb Tsipursky votes For

AI companies should be liable for harms caused by their deployed models

However, the problem with these proposals is that they require the coordination of numerous stakeholders from a wide variety of companies and government figures. Let me share a more modest proposal that’s much more in line with our existing methods o...

more AI Verified source (2023)

added 8mo ago