Singapore's agentic AI framework keeps a human in the loop: a memo of what your AI decided

What the framework says

In its parliamentary response of 5 August 2026, the Ministry of Digital Development and Information says the government released the Model Governance Framework for Agentic AI in January 2026. It emphasises human and organisational accountability, designated oversight roles, and controls appropriate to the risks and autonomy of the system.

The everyday version: review what the AI decided

Most people won't deploy an AI agent. But many will spend hours in a chat where the AI proposes, you agree, it revises, you change your mind. By the end, decisions have been made, some by you, some suggested by the AI and accepted without much thought.

Reviewing them means being able to answer:

• What's decided, and does it still stand?
• What changed along the way?
• What was set aside?
• What doesn't add up?

Review it by hand

1. Go through the chat and list every decision.
2. Mark which ones you made, and which ones you accepted from the AI.
3. Cross out what you later changed.
4. Look for anything that contradicts something else.

The faster way: Memo, in Uncook

Uncook's Memo reads the chat and gives you that review: what's decided, what changed along the way, what was set aside and what doesn't add up. On the ChatGPT page, click the Uncook icon and pick Memo.

The memo is for a person to read and check. It doesn't decide anything, and it doesn't make anyone compliant with any framework. The task cost is shown in the tool box below. Copy it, export it to PDF, Markdown, text or Word, or share it as a page.

Questions people ask

Does Uncook read AI agents' actions? No. It reads supported chats in desktop Chrome, not what an agent did in other tools.

Can I use the memo at work? Check your employer's rules on AI and confidential information first.

Does it work on several chats? Yes. Pick them together, and the memo shows where they agree, where they differ, and what's still open.