Microsoft Expands Copilot Studio With AI Agent and Governance Features

Arthur Kay
Updated •
AI, Microsoft

Microsoft rolled out several Copilot Studio updates in November 2025, largely shaped by announcements at Microsoft Ignite. The changes reinforce a clear direction: Copilot Studio is evolving from a workflow automation tool into a platform for building and governing AI agents inside organizations.

One of the most notable updates is the general availability of GPT-5 Chat in both the US and the EU. This removes a regional limitation that previously affected enterprise deployments and allows organizations to standardize agent behavior across markets.

Another key change is the introduction of human-in-the-loop (HITL) controls, now available in preview. HITL allows organizations to require human review or approval at specific stages of an agent's execution. This is aimed at reducing risk in scenarios where AI agents interact with sensitive data, trigger downstream actions, or make decisions that require accountability.

Microsoft is positioning these controls as a governance layer rather than a speed bump. Agents can still operate autonomously within defined boundaries, but escalation paths and checkpoints are built into the system by design.

Taken together, the November updates highlight a broader shift in how Copilot Studio is expected to be used. Instead of automating isolated steps, organizations are being encouraged to design systems where AI agents manage complexity, while humans provide judgment, validation, and oversight.

There are no major consumer-facing changes in this release. The focus remains squarely on enterprise use cases, compliance requirements, and large-scale operational workflows.

Comments

  1. JohnIL said on
    Reply

    AI won’t kill itself, but the companies trying to cram it down everyone’s PC’s and devices will. When you force people to adopt something many generally have a negative reaction. I have nothing against AI being used in areas benefiting humans. But I also think it has become this sort of technology that has been overly marketed like a Swiss army knife a do everything sort of tool.

  2. Mystique said on
    Reply

    I wonder if we could weaponize AI to attack itself by creating our own AI to find and reporting AI to be blocked in an large list to be used in an application like peerblock for system wide blocking and also a list for uBo. Blocking AI via IP and AI analytics would be wonderful.

    “Anti-AI bot has detected AI and has filtered and blocked it accordingly.”

    hahaha, I think its time we do.

    1. boris said on
      Reply

      Every shopping site chat assistance is AI bot. Do you want to go to any shopping site and never be able to resolve for example return problems?

      I would not mind blocking AI chat boxes on none-essential websites. But on every website? That ship has sailed. I would also prefer to use this list through a separate extension because it has to be turned on/turned off all the time.

    2. Arthur K said on
      Reply

      Hahaha

    3. random said on
      Reply

      I see what your idea is about but it will never work:
      These AI companies have complicit deals with ‘training data providers’ where the provider of all these scraped websites provides the data, and e.g. OpenAI implicitly avoids asking too many questions.

      These AI companies get their data to train their AI models from other companies that don’t respect the rules.

      They deliberately rent residential IP addresses from criminal botnets and sometimes from the Hola VPN network, then using a fake browser (headless Chrome with Chromedriver) they aggressively scrape your website while masquerading as legitimate users.

      They will use many multiple residential IP addresses at random (load balancing) so that each IP alone doesn’t send too many requests, and they will run these headless web browsers (automated) inside of virtual machines with common settings to defeat fingerprinting:
      – OS version, OS language, timezone, web browser and screen resolution are basically mimicking a real computer.

      The only way to defeat these AI bots is to use cryptographic proof-of-work systems.
      Basically encrypt a text with a random AES or Serpent key and ask the web browser to bruteforce the correct key then give the decrypted text.

      You do this using a Javascript code and by giving only part of the AES/Serpent key to make the bruteforce time reasonably short enough (e.g. give the first half of the AES-128 key and let the Javascript bruteforce the last 64-bits).

      This way it’s quick for your server but expensive for the web browser.

  3. Peter Newton said on
    Reply

    OMG ! LMAO ! Yes ! ,, you too can get data mined like never before ! DIE Monster DIE !

    Seriously ,, give them enough rope and they hang themselves LOL ! .. the software company and the fanboys and girls !

    Peter Newton [London UK Linux user since 2004]

    1. Allwynd said on
      Reply

      People are now joking that Windows is the best thing that could happen to Linux in a long time – they are making such bad decisions that more and more people are moving to Linux as we speak.

      I wouldn’t be surprised if Linux passes 5% world market in the desktop home user segment soon. And if this continues and Linux solidifies itself around the 10%, it will really make some waves.

  4. Allwynd said on
    Reply

    Amazing, Windows is getting better and more stable with every update. People are flocking from other OSes and leaving them behind to bask in the glory of Windows.

    1. Toxic said on
      Reply

      “Ray Liotta Laughing Meme”

Leave a Reply

By commenting, you accept our privacy policy and keep the discussion respectful and on topic.