Opinion: Microsoft’s brakes for AI agents are a good idea, and every agent platform should offer them
Risk scores, read-only modes and human approval steps make agents easier to trust. We think this kind of control should come built in, from every vendor.
Illustration · not a photoHow this story was checked
- ResearchedWith AI assistance · 3 sources
- AI source checkEach sentence matched to its source
- Editor approvalMayank Sahu · 20 Oct, 13:49
- Published20 Oct, 13:49
Verify this story
- 3 sources, 1 official · 3 with an archived copy · see sources
- Every fact has a label and a source number · fact ledger
- What we searched, read and could not verify · research note
- Machine-readable: facts as JSON · text as Markdown
How this was made: researched, drafted and source-checked by AI (Claude, made by Anthropic), then published under the editor’s standing approval. A person did not read it line by line before it went live. AI policy
In short
We think Microsoft’s new agent controls point in the right direction. Instead of an on-off switch, they slow an agent down step by step as its risk rises: first removing write access, then asking a person to approve, and only then stopping it. That is how careful teams already manage new staff, and it should become standard for every agent platform.
Why this approach works
Microsoft’s new controls give each AI agent a risk score and respond in steps: remove its ability to change things, then hold high-impact actions for human approval, then isolate or stop it.12 Microsoft’s security chief Hayete Gallot put the idea simply: “Model transparency will not replace containment.”2
We like this because it mirrors how good organisations treat a new employee. You do not hand over every key on day one, and you do not fire someone for one odd request. You watch, you limit, and you step in when needed. Graduated brakes let agents stay useful while risk is low.
Credit where it is due
Monitoring that sits outside the agent is a thoughtful design choice.2 An agent that goes wrong cannot simply switch off its own supervision. It is also helpful that Microsoft’s tools can recognise agents from other companies, such as Claude Code, OpenAI’s Codex and GitHub Copilot.2 Most workplaces use more than one vendor, so cross-vendor visibility matters.
What we would like to see next
- Clear pricing and availability. The controls are in private preview, and pricing has not been announced.3 Smaller companies will want to know if they can use them without the top-tier licence.
- Shared standards. If Google, AWS and others offer similar brakes, it would help if risk signals and approval steps worked the same way across platforms.
- Plain-language logs. When an agent is paused, the people involved should see why, in simple words.
What teams can do today
You do not need new products to start. Give each agent its own account with the least access it needs, require a human to approve anything that cannot be undone, and keep logs. These habits make any future tools work better.
Our view
Agents will do more real work each year. Building in brakes now is not a sign of distrust in AI; it is what makes wider use possible. Microsoft has offered a sensible template, and we hope others build on it.
Fact ledger
How labels workQuestions readers ask
Is this a news report?
No. This is Anawords’ opinion, based on our news report on Microsoft’s agent controls.
What can a small team do without these tools?
Give each agent its own limited account, require human approval for irreversible actions, and keep logs.