Microsoft first began rolling out Copilot connectors to Windows Insiders in October 2025, positioning them as a way to let the AI assistant search across personal services like OneDrive, Outlook, Google Drive, Gmail, Google Calendar, and Google Contacts using natural language.

These connectors are meant to give more contextual information to Copilot through various services, even third-party ones. Once enabled, Copilot can pull context from linked accounts to answer questions, locate files, find emails, or reference calendar items without you manually switching apps.

The Google Contacts connector, specifically, allows Copilot to retrieve saved contact details like phone numbers and email addresses directly from your Google account.

However, Google Contacts took its sweet time to be generally available to the public, but has now rolled out to everyone using Copilot web and the Copilot Android app. Some users have also received it in the standalone Copilot Windows app, but unfortunately, it’s not available in mine.

But I did get the connector for Google Contacts in both my Copilot Android app and web version, and curiosity got the best of me. I decided to enable it and see how well it actually works. To make things interesting, I also compared it with Google’s Gemini.

How to link Google Contacts to Microsoft Copilot

There are three ways to enable the Google Contacts connector in Copilot. I’ll tell the fancy one first:

Open a chat in Copilot web and literally ask it to check Google Contacts and tell the phone number of someone from your Google contact list.

Copilot will then tell you that it needs your consent to access Google Contacts. You can click the “Connect” button here, and Copilot will prompt you again, with the benefits and reassurances. Interestingly, Copilot allows you to connect to any Google account that you have, and it doesn’t have to be the one you’ve signed in with. Click the “Continue to Google Contacts” button to proceed.

You’ll then see the familiar “Sign in with Google” page. Since I have already enabled other Google connectors like Drive, Gmail, and Calendar, it shows Copilot needs “additional access” to my Google account.

This window also shows what permissions we have to give to Copilot, and unsurprisingly, giving access to Google Contacts will give Copilot the permission to not just see, but also download our contacts, which means that it would be up in Microsoft’s Cloud.

Click “Continue” and you’ll be back in the Copilot chat interface.

The second method is to enable the Google Contacts connector near the text field in Copilot. Click the connector icon and enable Google Contacts. You’ll get the same options to sign in with Google and give permissions to Copilot.

You can also enable Google Contacts connector from Copilot settings. Click your profile, select Settings, choose Connectors and then enable the toggle for Google Contacts.

For each connector from Google, you’ll have to sign in separately with Google. As of writing, only the Copilot web has got access to Google Contacts. The standalone Copilot app will gradually get it, though.

Testing Google Contacts connector in Microsoft Copilot

Microsoft’s official support document notes that you can only retrieve contact information using connectors, such as “What is John’s email address?”

That’s it. You cannot create new contacts or edit existing contacts. You can’t even ask Copilot to send a text message to a particular Google Contact.

About a year ago, the Copilot app got Phone Connection, with which you could set alarms and timers, and view and send text messages. That is a completely different feature that connects Copilot with Phone Link, but in our testing during that time, it was a bit finicky and not everything worked as expected.

However, Copilot was able to fetch the phone number correctly using Phone Connection.

This was before connectors. Doing the same now would show the contact info of the person from Outlook, since I have already synced my contacts to Outlook/OneDrive, and the Connection too is enabled.

Now let’s see if the connectors for Google Contacts work as expected, because not all people enabled Phone Connection, and not everyone syncs their contacts with OneDrive. People with Android smartphones already have their contacts synced with Google.

Earlier, I asked Copilot to fetch me Mayank’s phone number from Google Contacts, which was when it asked me to give permission. After enabling all permissions, I tried again, but Copilot was unable to fetch the number, saying that there was no entry under the same name.

Then I tried another contact, and Copilot gave me the right response, which prompted me to prompt (pun intended) to fetch another contact’s number, and Copilot worked here as well. It even got the pronouns right, which I have no idea how it did.

I again asked Copilot to fetch a fourth contact and a fifth contact. One worked while the other didn’t.

But this time, I understood why Copilot was unable to show the phone numbers of some contacts. For some reason, Copilot can show a contact’s phone number only if that person’s email ID is added in the contact info. The first contact’s phone number I asked for was Mayank, and his email ID wasn’t added to the contact info that I saved in Google Contacts. I tried the prompt again after adding his email ID and guess what…

After adding his email ID, Copilot shows phone number and email ID, with a Google logo, confirming that the info was indeed taken from Google Contacts.

I confirmed it with a few other contacts as well. Copilot shows the phone number only if they have email IDs stored in the contact details. Although I couldn’t find an official explanation for this odd behaviour, my theory is that Microsoft did this for privacy reasons (remember this, as it would come in handy in just a few moments).

Comparing Copilot Connectors with Google Gemini

While Microsoft’s Copilot has its connectors, Google’s Gemini has Connected Apps, and if social media is proof of anything, it’s that Copilot is far behind when compared with Gemini. However, it was Microsoft that started early in the AI race to make an “Agentic OS” and this head start has definitely got some upsides to it, despite the general public despising the company’s AI efforts.

Gemini is truly a more advanced AI model when compared, and there is no doubt about it, but it isn’t as good as Copilot when it comes to a simple task of fetching a contact’s phone number or even sending them an email.

For context, I have a Gemini PRO subscription, and its Connected Apps toggle is already turned on for Google Workspace, which includes Gmail, Google Drive, and the lot.

As for Copilot, I am not a paid subscriber, for obvious reasons, and it’s on the free tier.

I asked the same prompt I gave to Copilot, hoping that Gemini would easily fetch Mayank’s phone number, considering that Google Contacts belongs to Google, and that I also added Mayank’s email ID to Google Contacts.

However, I was surprised to see that Gemini was unable to retrieve his phone number. To be fair to Google, their Workspace doesn’t explicitly say anything about including “Google Contacts”. But I digress, as both of these are Google products, and I was half expecting Gemini to say something like “this is Mayank’s phone number, do you want me to send a text message?”

Of course, Copilot can’t send a text directly, and it can only retrieve phone numbers and email IDs. However, it does have the ability to draft an email and, with one click, open Outlook to send an email.

Naturally, one would expect Gemini to do even better, considering it already connects to Gmail and that it is a PRO subscription. Also, note that I have already sent an email to Mayank from Gmail, so retrieving it could be a piece of cake for Gemini, but I was wrong.

Gemini, despite having direct access to the Gmail workspace, was unable to give me a link that would automatically add the subject and body. In fact, it didn’t give any link at all. And I can’t be fair with Google here, because at first it wasn’t even capable of finding Mayank’s email ID.

It was only after I sent an email to Mayank in Gmail that Gemini was able to fetch his email. Still, it was unable to find the phone number or give me an easy link to send him an email.

Copilot, despite being in the free tier, was able to use its connectors to find phone numbers and email addresses and also craft an email with a link that directly opened Outlook with the subject and body already intact. Isn’t this what AI is supposed to do?

Don’t get me wrong, I’m not praising Microsoft for all the AI that they have poured over in Windows 11, but credit where it’s due, the company’s head start in AI and aggressive push towards an Agentic OS did bring about a system that behaves more like an agentic AI and not a mere large language model.

That being said, I’m also aware of the agentic features coming to Android soon, but Copilot isn’t far either, as Microsoft announced Copilot Tasks, which is a proper agentic mode that interprets your prompts and makes them into step-by-step tasks and executes them in the background.

Sure, the hate around Copilot is understandable considering how aggressive Microsoft was with the branding, and throwing away every bit of logic to slap the Copilot name and logo throughout Windows. But in the end, an AI assistant is only as useful as the amount of data about us that it has. In that case, Copilot comes on top, even above Gemini.

The privacy-conscious people wouldn’t go anywhere near Copilot, but what they fail to realise is that Agentic features, even the ones coming to Android, iOS, and macOS, all require explicit access to your data.

Nevertheless, if you do not want multi-trillion-dollar companies to tell your contact information, you can always turn it off. And Copilot connectors, too, are just a toggle away.

The post Tested: Microsoft Copilot can now access your Google Contacts, and it’s better than Gemini integration appeared first on Windows Latest

On Friday afternoon, Donald Trump posted on Truth Social, accusing Anthropic, the AI company behind Claude, of attempting to "STRONG-ARM" the Pentagon and directing federal agencies to "IMMEDIATELY CEASE" use of its products. At issue is Anthropic CEO Dario Amodei's refusal of an updated agreement with the US military agreeing to "any lawful use" of Anthropic's technology, as Defense Secretary Pete Hegseth mandated in a January memo, to the frustration of many tech workers across the industry.

As we explained earlier this week, that agreement would give the US military access to use the company's services for mass domestic surveillance and …

Read the full story at The Verge.

Source

Red Hat has been deploying AI in the enterprise for some time. For example, Red Hat Enterprise Linux (RHEL) now comes ready to do AI work; Red Hat has its own inference engine; and the company offers AI-enabled sysadmin tools.

Now, Red Hat is positioning itself as a full-stack AI platform vendor. The IBM-owned Linux powerhouse is doing this with a new Red Hat AI Enterprise (RHAE) suite and a jointly engineered Red Hat AI Factory with NVIDIA.

Red Hat AI Enterprise

RHAE is a new integrated AI platform designed to deploy and manage models, agents, and applications across hybrid environments using RHEL and OpenShift as its foundation. Red Hat pitches it as a “metal-to-agent” stack that unifies the AI lifecycle, enabling IT to manage AI like any other enterprise system rather than isolated pilots that never reach production.

The platform bundles high-performance inference, model tuning and customization, and agent deployment and management, with support for “any model, any hardware, any environment” as long as it can be layered on top of Red Hat’s Linux and Kubernetes platforms. Red Hat argues that this approach offers a more consistent, security-hardened environment for running AI at scale, using the same tools and processes enterprises already have around OpenShift.

This is all part of the usual Red Hat integrated stack approach.

Alongside AI Enterprise, Red Hat is shipping Red Hat AI 3.3. The release adds compressed, production-ready versions of models such as Mistral-Large-3, Nemotron-Nano, and Apertus-8B-Instruct via the OpenShift AI Catalog. It also supports newer models like Mistral 3 and DeepSeek-V3.2, as well as multimodal upgrades, including faster Whisper speech processing and enhanced tool calling for agentic workflows.

AI 3.3 also introduces a technology preview of models-as-a-service, giving internal users self-service access to privately hosted models via an API gateway, standardizing how AI is consumed within large organizations.

On the hardware side, Red Hat is broadening support to include generative AI on Intel CPUs for small language models, expanding certification for NVIDIA’s Blackwell Ultra GPUs and AMD’s MI325X accelerators, and adding internal GPU-as-a-service features such as automatic checkpointing to prevent long-running jobs from losing work.

To address governance and software supply chain concerns, Red Hat is adding a new AI Python Index as a trusted repository of hardened versions of key AI tools. These tools include Docling, an IBM-developed open-source toolkit for converting unstructured documents into machine-readable formats for model training, and SDG Hub, a framework for building synthetic data generation pipelines used to fine-tune large language models.

The company is also tightening AI observability and safety with more detailed telemetry across workloads and a tech preview of integrated NVIDIA NeMo Guardrails to enforce policy and alignment in AI interactions.

Red Hat AI Factory

This is a good time to also introduce the Red Hat AI Factory with NVIDIA. This is a direct follow-up to Red Hat’s recent release of a customized RHEL for NVIDIA’s Vera Rubin AI Platform. This new co-engineered software platform partners RHAE up with NVIDIA AI Enterprise to create what the companies describe as an end-to-end AI stack optimized for large-scale deployments. 

The AI Factory is aimed squarely at enterprises trying to move from ad hoc AI projects to “industrial-scale” production systems. It’s meant to help operations teams manage both conventional infrastructure and AI-specific demands, from provisioning and GPU orchestration to model performance and security, under a single umbrella.

This joint platform focuses on three main themes: time-to-value, performance and cost, and enterprise security posture. For faster deployment, customers get streamlined workflows and immediate access to pre-configured models, including IBM’s indemnified Granite family as well as NVIDIA Nemotron and NVIDIA Cosmos open models delivered via NVIDIA NIM microservices, with NVIDIA NeMo available for enterprise data tuning.

On the performance side, the AI Factory relies on Red Hat’s inference stack, powered by vLLM, the popular open-source inference engine, and NVIDIA technologies such as TensorRT-LLM, Dynamo, and the BlueField data processing unit. This is coupled with built-in observability to help organizations meet strict service-level objectives and reduce AI total cost of ownership. 

Chris Wright, Red Hat CTO, describes the AI Factory with NVIDIA as part of a broader shift from AI experimentation to “industrial-scale, enterprise-wide production,” arguing that managing the entire AI computing stack with the same rigor as core IT platforms is now a necessity for large organizations. In both cases, Red Hat is betting that its long-standing expertise in hybrid cloud infrastructure gives it an edge as enterprises seek to standardize AI across multiple environments.

Red Hat is also lining up ecosystem support, with players such as Cisco, Dell, Lenovo, Supermicro, TD SYNNEX, and WWT signaling plans to deliver validated infrastructure and channel offerings around AI Enterprise and the AI Factory. Their role is to package Red Hat’s software with AI-optimized hardware and services so customers can treat AI as a standard enterprise workload rather than a bespoke science project. So, if you like the idea of a single AI software stack based primarily on NVIDIA hardware, boy, does Red Hat want to talk to you. 

The post Red Hat introduces its first out and out AI platform appeared first on The New Stack.

Amazon Web Services (AWS) is announcing two new features for the AWS Command Line Interface (AWS CLI) v2: structured error output and the “off” output format.

Structured error output

Errors returned from AWS service APIs often include useful details beyond the code and message—bucket names, validation reasons, resource IDs—that were previously hidden unless you used --debug. Now, you can see this error information directly in your error output.

Starting with AWS CLI v2 version 2.34.0, any additional error details returned from service APIs will now be shown in the stderr output. Additionally, you can configure the AWS CLI to output your errors in alternative structured formats. Control how errors are displayed using the new --cli-error-format CLI flag, the cli_error_format configuration setting, or the AWS_CLI_ERROR_FORMAT environment variable.

Supported formats for the error format parameter:

• enhanced (default) – Error message with additional details displayed inline
• json – JSON structure with all error fields
• yaml – YAML structure with all error fields
• text – Tab-delimited error fields
• table – ASCII table format
• legacy – Original error format

Accessibility enhancements

Since September 2025, AWS CLI errors started including the aws: [ERROR]: prefix for some exceptions. This prefix signals that an error has occurred and supports accessibility best practices and automation use cases. This release ensures the prefix is consistently included for all errors in the enhanced and legacy formats.

Example: Using enhanced output error format

$ aws s3api get-object \
    --bucket amzn-s3-demo-bucket \
    --key file.txt out.txt \
    --cli-error-format enhanced

aws: [ERROR]: An error occurred (NoSuchBucket) when calling the GetObject operation: The specified bucket does not exist

Additional error details:

BucketName: amzn-s3-demo-bucket

Example: Using json output error format

$ aws s3api get-object \
    --bucket amzn-s3-demo-bucket \
    --key file.txt out.txt \
    --cli-error-format json
{
  "Code": "NoSuchBucket",
  "Message": "The specified bucket does not exist",
  "BucketName": "amzn-s3-demo-bucket"
}

Example: Using legacy output error format

$ aws s3api get-object \
    --bucket amzn-s3-demo-bucket \
    --key file.txt out.txt \
    --cli-error-format legacy

aws: [ERROR]: An error occurred (NoSuchBucket) when calling the GetObject operation: The specified bucket does not exist

Turning off CLI output

Sometimes, you might want to hide the AWS CLI command output, such as when using a command that may output sensitive information. The off format suppresses stdout while still preserving errors on stderr.

For example, you can create an AWS Secrets Manager secret without writing the secret ARN or version information to logs:

$ aws secretsmanager create-secret \
    --name my-secret \
    --secret-string "password123" \
    --output off
$ echo $?
0

You can set this using the--output off CLI flag, setting output = off in your configuration file, or the AWS_DEFAULT_OUTPUT=off environment variable.

Next Steps

To take advantage of these new output features, upgrade your version of the AWS CLI to 2.34.0. For more information, see the Structured error output and Output format guides. Please share your questions, comments, and issues with us on GitHub.