Microsoft Intros’Cowork’Feature for Copilot, AI Updates
Microsoft has actually introduced a trio of AI updates covering Microsoft 365 Copilot, Security Copilot, and Microsoft Foundry.
The top news was the statement of Copilot Cowork, a new mode in Microsoft 365 Copilot that Microsoft says is created to help users hand off multistep tasks instead of merely getting the answer back in chat.
“Copilot Cowork is developed for that: It helps Copilot act, not just chat,” said Charles Lamanna, president of Organization Applications and Representatives.
Copilot Cowork Moves Beyond Chat
Cowork is developed to take a user’s goal and turn it into a structured strategy that runs in the background. Microsoft stated it pulls context from across Microsoft 365(Outlook, Teams, Excel, files, and meetings) through what it calls Work IQ, then surface areas checkpoints for approval before making changes.
According to Microsoft, Cowork can manage tasks such as dealing with calendar conflicts, preparing meeting briefs, putting together research study memos with citations from web and work environment sources, and more.
Microsoft said Cowork operates within the existing Microsoft 365 security and governance structure, with identity, permissions and compliance policies imposed by default. Actions and outputs are auditable, the company said.
Microsoft also stated Cowork can tap Claude from Anthropic, which Lamanna referred to as a “multi-model advantage” that enables Copilot to path work to the design best suited for the job.
Cowork is presently readily available to a limited set of clients through a Research Sneak peek. Microsoft said it expects more comprehensive rollout through its Frontier program in late March 2026. The business presented Frontier previously this year as an early-access channel for emerging Copilot functions.
Security Copilot Gets a Credential-Finding Upgrade
Microsoft also revealed the basic accessibility of Agentic Trick Finder, or ASF, in Microsoft Security Copilot.
The function is focused on spotting exposed qualifications concealed in disorganized data such as emails, chat logs, documents and screenshots. Microsoft said ASF uses a multi-step, multi-agent reasoning procedure to determine whether a suspicious string is a valid credential and what level of access that credential might provide.
“Unlike regex-based scanners, ASF utilizes reasoning to identify not just qualifications, however the systems they unlock, assisting security teams understand direct exposure and react faster,” Microsoft composed in the statement.
Microsoft said that technique is meant to enhance triage by decreasing the incorrect positives typically created by standard pattern-matching tools, while also recognizing qualifications that do not fit recognized formats.
In benchmark screening using artificial datasets throughout e-mails, chats, notes, and files, Microsoft stated ASF reached 98.33% credential recall with no false positives. Standard regex-based tools, the company stated, identified about 40% of the very same credentials.
ASF currently supports more than 20 credential types, including Azure Storage Keys, AWS Access Keys, OAuth tokens, SSH personal secrets, and database connection strings. Microsoft stated it is likewise checking out GitHub integration to extend the capability into source code analysis.
Fireworks AI shows up in Foundry
The company’s 3rd announcement was a public sneak peek that brings Fireworks AI to the Microsoft Foundry model brochure.
Microsoft said the integration provides developers access to Fireworks AI’s cloud-based reasoning engine inside their Foundry tasks, using low-latency inference for a number of open source models.
“For customers requiring the latest open source models from emerging frontier laboratories, break-neck speed, or the capability to deploy their own post-trained custom-made models, Fireworks delivers best-in-class inference efficiency,” Microsoft said in the announcement.
At launch, the sneak peek supports both serverless pay-per-token releases and provisioned throughput throughout 4 models: Minimax M2.5, OpenAI’s gpt-oss-120b, MoonshotAI’s Kimi-K2.5, and DeepSeek-v3.2.
Microsoft said clients can also import and deploy their own fine-tuned versions from those model households– consisting of Qwen3-14B and DeepSeek v3.1– through a brand-new Custom-made Models workflow in Foundry.
The Fireworks combination is opt-in during preview and must be enabled through the Azure website’s Preview functions panel. Microsoft stated clients likewise must remain in among 6 supported U.S. areas to use the pay-per-token alternative.