Windows 10: Microsoft begins preparing Windows 11 for its agentic AI future

GHacks · Nov 18, 2025 at 7:12 AM

Microsoft released a new insider preview build to the developer and beta channels of Windows 11. The new build 26220.7262 introduces several new features and changes, including a new "experimental agentic features" toggle in the Settings.

If you recall, Microsoft announced this month that it is turning Windows into an agentic operating system. The post, by Windows-chief Pavan Davuluri on X, faced heavy criticism. To the point that Davuluri limited the commenting ability.

The new AI component setting in the development edition of Windows 11 is a major step towards that goal. First, the good news. Microsoft is shipping the feature turned off. Windows users who want to enable the experimental AI features need to turn it on first before they can do so.

Here is what happens when you activate the toggle: Only administrators may activate the feature, but it will then be enabled for all users on the system automatically.

Once enabled, three main and far-reaching changes occur on the system that give AI access to a wider range of features.

Windows may create AI agents with separate accounts on the device, that act on the user's behalf.

Creation of agent workspaces, that allow them to "work in parallel with a human user". This happens using runtime isolation and scoped authorization, according to Microsoft.

Agentic apps may request and access six folders: Documents, Downloads, Desktop, Music, Pictures and Videos.

Microsoft is introducing several new key concepts into Windows that may require an explanation.

A key component for Windows' agentic future is the agent workspace. Think of it as a separate isolated desktop for AI agents. This separates the AI from the user's own desktop, but still allows for some interactions to occur.

Notably, agentic AIs have access to all apps by default that all users of the operating system have access to. If an app is installed system-wide, the AI has access to it as well.

Microsoft says that access can be limited by installing an app only for a specific user. Also, the ability to install apps for specific AI agents is also coming.

Besides access to apps, agentic AI is also getting file access. Access is limited to the six folders mentioned above and all folders that all "authenticated users have access to".

Agentic security principles

Microsoft recommends that administrators enable the experimental feature only if they are aware of certain security implications:

AI models face functional limits in terms of how they behave.

Hallucination, which means made-up information, is still a problem.

Agentic AI may also introduce new security risks, like cross-prompt injection.

The implementation brings along with it "robust security and privacy controls that empower customers to explore their potential confidently with the support of clear guidance and appropriate guardrails driven by these goals", according to Microsoft.

Three goals are mentioned specifically:

Non-repudiation: Which should make all agent actions "observable and distinguishable".

Confidentiality: Agents that access user data should "meet or exceed the security and privacy standards of the data".

Authorization: The user approves "all queries for user data as well as actions taken".

Microsoft has not revealed when the new functionality will become available in stable versions of the Windows operating system. It seems likely that this won't happen for quite a while, considering the experimental nature of the feature and the delicate nature of introducing agentic AI into the operating system. My guess? Could be part of Windows 11, version 26H2, the main feature update for the operating system in 2026.

Now You: what is your take on agentic AI? Something that you use already, plan to use, or have no need for? Feel free to leave a comment down below.

Thank you for being a Ghacks reader. The post Microsoft begins preparing Windows 11 for its agentic AI future appeared first on gHacks Technology News.

read more...

Brink · Nov 18, 2025 at 7:15 AM

Securing the future of AI and machine learning at Microsoft

Artificial intelligence (AI) and machine learning are making a big impact on how people work, socialize, and live their lives. As consumption of products and services built around AI and machine learning increases, specialized actions must be undertaken to safeguard not only your customers and their data, but also to protect your AI and algorithms from abuse, trolling, and extraction.

We are pleased to announce the release of a research paper, Securing the Future of Artificial Intelligence and Machine Learning at Microsoft, focused on net-new security engineering challenges in the AI and machine learning space, with a strong focus on protecting algorithms, data, and services. This content was developed in partnership with Microsoft’s AI and Research group. It’s referenced in The Future Computed: Artificial Intelligence and its role in society by Brad Smith and Harry Shum, as well as cited in the Responsible bots: 10 guidelines for developers of conversational AI.

This document focuses entirely on security engineering issues unique to the AI and machine learning space, but due to the expansive nature of the InfoSec domain, it’s understood that issues and findings discussed here will overlap to a degree with the domains of privacy and ethics. As this document highlights challenges of strategic importance to the tech industry, the target audience for this document is security engineering leadership industry-wide.

Our early findings suggest that:

1. Secure development and operations foundations must incorporate the concepts of Resilience and Discretion when protecting AI and the data under its control.

AI-specific pivots are required in many traditional security domains such as Authentication, Authorization, Input Validation, and Denial of Service mitigation.

Without investments in these areas, AI/machine learning services will continue to fight an uphill battle against adversaries of all skill levels.

2. Machine learning models are largely unable to discern between malicious input and benign anomalous data. A significant source of training data is derived from un-curated, unmoderated public datasets that may be open to third-party contributions.

Attackers don’t need to compromise datasets when they are free to contribute to them. Such dataset poisoning attacks can go unnoticed while model performance inexplicably degrades.

Over time, low-confidence malicious data becomes high-confidence trusted data, provided that the data structure/formatting remains correct and the quantity of malicious data points is sufficiently high.

3. Given the great number of layers of hidden classifiers/neurons that can be leveraged in a deep learning model, too much trust is placed on the output of AI/machine learning decision-making processes and algorithms without a critical understanding of how these decisions were reached.

AI/machine learning is increasingly used in support of high-value decision-making processes in medicine and other industries where the wrong decision may result in serious injury or death.

AI must have built-in forensic capabilities. This enables enterprises to provide customers with transparency and accountability of their AI, ensuring its actions are not only verifiably correct but also legally defensible.

When combined with data provenance/lineage tools, these capabilities can also function as an early form of “AI intrusion detection,” allowing engineers to determine the exact point in time that a decision was made by a classifier, what data influenced it, and whether or not that data was trustworthy.

Our goal is to bring awareness and energy to the issues highlighted in this paper while driving new research investigations and product security investments across Microsoft. Read the Securing the Future of Artificial Intelligence and Machine Learning at Microsoft paper to learn more.
Click to expand...

Source: Securing the future of AI and machine learning at Microsoft - Microsoft Secure

Brink · Nov 18, 2025 at 7:15 AM

Securing the future of AI and machine learning at Microsoft

Artificial intelligence (AI) and machine learning are making a big impact on how people work, socialize, and live their lives. As consumption of products and services built around AI and machine learning increases, specialized actions must be undertaken to safeguard not only your customers and their data, but also to protect your AI and algorithms from abuse, trolling, and extraction.

We are pleased to announce the release of a research paper, Securing the Future of Artificial Intelligence and Machine Learning at Microsoft, focused on net-new security engineering challenges in the AI and machine learning space, with a strong focus on protecting algorithms, data, and services. This content was developed in partnership with Microsoft’s AI and Research group. It’s referenced in The Future Computed: Artificial Intelligence and its role in society by Brad Smith and Harry Shum, as well as cited in the Responsible bots: 10 guidelines for developers of conversational AI.

This document focuses entirely on security engineering issues unique to the AI and machine learning space, but due to the expansive nature of the InfoSec domain, it’s understood that issues and findings discussed here will overlap to a degree with the domains of privacy and ethics. As this document highlights challenges of strategic importance to the tech industry, the target audience for this document is security engineering leadership industry-wide.

Our early findings suggest that:

1. Secure development and operations foundations must incorporate the concepts of Resilience and Discretion when protecting AI and the data under its control.

AI-specific pivots are required in many traditional security domains such as Authentication, Authorization, Input Validation, and Denial of Service mitigation.

Without investments in these areas, AI/machine learning services will continue to fight an uphill battle against adversaries of all skill levels.

2. Machine learning models are largely unable to discern between malicious input and benign anomalous data. A significant source of training data is derived from un-curated, unmoderated public datasets that may be open to third-party contributions.

Attackers don’t need to compromise datasets when they are free to contribute to them. Such dataset poisoning attacks can go unnoticed while model performance inexplicably degrades.

Over time, low-confidence malicious data becomes high-confidence trusted data, provided that the data structure/formatting remains correct and the quantity of malicious data points is sufficiently high.

3. Given the great number of layers of hidden classifiers/neurons that can be leveraged in a deep learning model, too much trust is placed on the output of AI/machine learning decision-making processes and algorithms without a critical understanding of how these decisions were reached.

AI/machine learning is increasingly used in support of high-value decision-making processes in medicine and other industries where the wrong decision may result in serious injury or death.

AI must have built-in forensic capabilities. This enables enterprises to provide customers with transparency and accountability of their AI, ensuring its actions are not only verifiably correct but also legally defensible.

When combined with data provenance/lineage tools, these capabilities can also function as an early form of “AI intrusion detection,” allowing engineers to determine the exact point in time that a decision was made by a classifier, what data influenced it, and whether or not that data was trustworthy.

Our goal is to bring awareness and energy to the issues highlighted in this paper while driving new research investigations and product security investments across Microsoft. Read the Securing the Future of Artificial Intelligence and Machine Learning at Microsoft paper to learn more.
Click to expand...

Source: Securing the future of AI and machine learning at Microsoft - Microsoft Secure

Brink · Nov 18, 2025 at 7:15 AM

Announcing future user-agents for Bingbot

As announced in October, Bing is adopting the new Microsoft Edge as the engine to run JavaScript and render web pages. We have already switched to Microsoft Edge for thousands of web sites “under the hood”. This evolution was transparent for most of the sites and we carefully tested to check whether each website is rendering fine on switching to Microsoft Edge. Over the coming months, we will scale this migration to cover all the sites.

So far, we were crawling using an existing bingbot user-agents. With this change, we will start the transition to a new bingbot user-agent, first for sites which require it for rendering and then gradually and carefully to all sites.

Bingbot user-agents today

Mozilla/5.0 (compatible; bingbot/2.0; +Which Crawlers Does Bing Use? - Bing Webmaster Tools)

Mozilla/5.0 (iPhone; CPU iPhone OS 7_0 like Mac OS X) AppleWebKit/537.51.1 (KHTML, like Gecko) Version/7.0 Mobile/11A465 Safari/9537.53 (compatible; bingbot/2.0; +Which Crawlers Does Bing Use? - Bing Webmaster Tools)

Mozilla/5.0 (Windows Phone 8.1; ARM; Trident/7.0; Touch; rv:11.0; IEMobile/11.0; NOKIA; Lumia 530) like Gecko (compatible; bingbot/2.0; +Which Crawlers Does Bing Use? - Bing Webmaster Tools)

In addition to the existing user-agents listed above, following are the new evergreen Bingbot user-agents

Desktop

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; bingbot/2.0; +Which Crawlers Does Bing Use? - Bing Webmaster Tools) Chrome/W.X.Y.Z Safari/537.36 Edg/W.X.Y.Z

Mobile

Mozilla/5.0 (Linux; Android 6.0.1; Nexus 5X Build/MMB29P) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/W.X.Y.Z Mobile Safari/537.36 Edg/W.X.Y.Z (compatible; bingbot/2.0; +Which Crawlers Does Bing Use? - Bing Webmaster Tools)
We are committing to regularly update our web page rendering engine to the most recent stable version of Microsoft Edge thus making the above user agent strings to be evergreen. Thus, "W.X.Y.Z" will be substituted with the latest Microsoft Edge version we're using, for example “80.0.345.0".

How to test your web site

For most web sites, there is nothing to worry as we will carefully test the sites to dynamically render fine before switching them to Microsoft Edge and our new user-agent.

We invite you to install and test the new Microsoft Edge to check if your site looks fine with it. If it does then you will not be affected by the change. You can also register your site on Bing Webmaster Tools to get insights about your site, to be notified if we detect issues and to investigate your site using our upcoming tools based on our new rendering engine.

We look forward to sharing more details in the future.

Thanks,
Fabrice Canel
Principal Program Manager
Microsoft - Bing
Click to expand...

Source: Announcing future user-agents for Bingbot | Webmaster Blog

Windows 10: Microsoft begins preparing Windows 11 for its agentic AI future

Microsoft begins preparing Windows 11 for its agentic AI future

Microsoft begins preparing Windows 11 for its agentic AI future

Microsoft begins preparing Windows 11 for its agentic AI future

Microsoft begins preparing Windows 11 for its agentic AI future - Similar Threads - Microsoft begins preparing

Finally! Microsoft is making AI useful in Windows by introducing AI agents

How to create Copilot agents using Microsoft AI?

Windows update prepares,begins to install and then goes back to preparing

Windows update prepares,begins to install and then goes back to preparing

Announcing future user-agents for Bingbot

Securing the future of AI and machine learning at Microsoft

Advancing the future of society with AI and the intelligent edge

Preparing for a Future Free Upgrade

Microsoft shares its vision of the future