Skip to content
Agents of Humanity
Public preview. Results are preliminary (criteria v1.0, 30 Sept 2026). Tell us what we got wrong
Autonomous agenthostedProprietary

Microsoft Copilot Cowork

Microsoft · microsoft.com/en-us/copilot/features/cowork

Copilot Cowork is a hosted agent in Microsoft 365 that carries out multi-step tasks (email, calendar, documents, Teams) using the user's Microsoft 365 account. It runs on models that Microsoft chooses, including Anthropic models as a subprocessor. It offers per-action approval prompts, pause and cancel controls, owner-authored Markdown skills in OneDrive, and Purview audit records. The agent cannot be exported, run elsewhere or independently verified, and Microsoft controls its model, base instructions and built-in skills.

Strengths
  • Approval prompts before sending, posting or scheduling, with previews and parameter view (C2 partial)
  • Custom skills are plain Markdown files in the owner's OneDrive (M1 partial)
  • Contractual 90-day data extraction window after subscription expiry (P5 partial)
  • Prompts and responses are documented as not used to train foundation models (C4 partial)
Gaps
  • No export of the agent and no way to run it outside Microsoft (P1, P2, P3 fail)
  • Model is chosen by Microsoft and not shown to the owner, and the base prompt is hidden (P4, T2, M5 fail)
  • Audit records have no owner-checkable tamper evidence and are kept for 180 days by default (A1, A2 fail)
  • No owner-set limits on instruction sources or recipients, and no single-step deletion of the whole agent (C1, C6 fail)
Evidence

All 26 findings

Hosted SaaS (web at m365.cloud.microsoft, plus the Microsoft 365 Copilot desktop app). The Cowork docs read are from MicrosoftDocs/microsoft-365-docs at commit eab9d7696cdff87474698b08a1fb328091102a2f (Cowork pages dated 2026-03-30 to 2026-04-22, labelled Frontier preview). The product page says Cowork is sold with usage-based billing. Reviewed 2026-09-30. Closed source; no hands-on testing.

Portable · Can you leave, take the whole agent with you, and keep it alive?

10%
  • P1
    Owner-usable round trip

    Cowork documents no export of the agent and no import or restore on any runtime Microsoft does not operate. Only individual output files can be downloaded (singly or as a zip of up to 50 files / 500 MB).

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Offer an agent export plus a matching importer or open runtime that restores it outside Microsoft
    Fail
  • P2
    Complete export

    Only created files and custom SKILL.md files (both in OneDrive) can be taken out by the owner. Task conversations, scheduled prompts, 'don't ask again' permissions, built-in skills and the base configuration stay with Microsoft, and there is no owner-facing agent export.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Export every class of agent state (tasks, schedules, permissions, instructions, history) in one bundle
    Fail
  • P3
    Runs without the provider

    Cowork runs only as a Microsoft-hosted service. It needs an active Microsoft 365 Copilot licence, Frontier enrolment and Microsoft accounts. Microsoft may change or discontinue previews at any time without notice.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Provide a runtime the owner can run indefinitely without Microsoft accounts, licences or services
    Fail
  • P4
    Model independence

    Microsoft selects the model ('Cowork picks the model best suited for the job'), and Cowork requires Anthropic to be enabled as a Microsoft subprocessor. There is no option to choose another model or a local one.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Let the owner configure any model meeting a documented minimum, including a local open-weight model
    Fail
  • P5
    Exit survives termination

    The Product Terms give 90 days to extract customer data (including OneDrive files and skills) from a disabled account after expiry. Public evidence does not settle whether for-cause termination is covered, and previews can be discontinued without notice and with no published shutdown or export plan.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Commit in binding terms to an export window that also covers bans and discontinuation of Cowork
    Partial

Transparent · Can you see everything the agent is, and everything shaping it?

25%
  • T1
    Readable without the provider

    Task state, schedules, permissions, built-in skills and the base logic live in Microsoft's service and can be read only through Microsoft's interfaces (the Cowork UI, Purview, eDiscovery). Microsoft holds the default encryption keys. Only custom skills are plain Markdown.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Store agent state in documented formats the owner can read without Microsoft services
    • Let the owner hold the keys
    Fail
  • T2
    What runs is what you see

    The base prompt and assembled context are not shown to the owner, and some background operations such as script execution run without showing their steps. The model in use cannot be determined: the Purview audit docs state 'Currently, Cowork doesn't show provider details'.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Show the full assembled context for any call
    • Show the model and version in effect for each run
    Fail
  • T3
    No shadow data

    Microsoft publishes bounded maximum retention for Microsoft 365 data classes (30 days after active deletion, 180 days after the subscription ends) and admins can search Copilot interactions. Microsoft-held service data (identifiers, audit records, execution-environment data) is not visible item by item, and Cowork-specific retention is not documented separately.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Publish a Cowork retention table covering the execution environment and subprocessor data
    • Show the owner every item Microsoft retains
    Partial
  • T4
    No hidden steering

    No paid placement is documented, and Cowork openly works through Microsoft 365 apps and Microsoft-chosen models. However, these first-party defaults cannot be switched off or replaced, and no binding commitment against self-preferencing was found.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Put a no-steering commitment in binding terms
    • Make first-party tool and model defaults replaceable
    Partial

Auditable · Can you reconstruct exactly what the agent did, and trust the record?

17%
  • A1
    Complete action record

    Purview CopilotInteraction records list accessed resources with read/create/modify actions, but not action arguments such as the email content sent. Audit Standard keeps them for 180 days. Coverage of scheduled runs is not documented, and there is no documented notice to the owner when recording fails.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Record arguments and targets for every action on every path, including schedules
    • Keep the record until the owner deletes it and alert on recording failure
    Fail
  • A2
    Tamper-evident, owner-checkable

    Microsoft holds the Purview audit log, and no hash chain, signature or independent verifier is documented that would let the owner detect edits or deletions.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Publish integrity data anchored with the owner or an independent witness, plus an open verifier
    Fail
  • A3
    Agent cannot rewrite its record

    Microsoft's service writes the audit records and Cowork's documented skills have no audit-log access. However, no protection against the agent altering or suppressing records is documented, and Cowork acts with the user's (here, admin's) permissions and runs scripts in its sandbox.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Document that the agent cannot reach, alter or suppress its audit records, and how the owner can check this
    Partial

Verifiable · Can you prove what it runs, who it is, and what changed?

17%
  • V1
    Obtainable source, matching build

    Cowork and its required Microsoft 365 services are closed source.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Publish the source of every required component under terms allowing modification and third-party maintenance
    • Make builds verifiable
    Fail
  • V2
    Owner-rooted identity

    Cowork acts as the user's Microsoft Entra / Microsoft 365 account ('Every action Cowork takes is authorized through your Microsoft 365 account'). That identity is issued by Microsoft and Microsoft can revoke it.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Root agent identity in an owner-held key or domain that recipients can verify without Microsoft
    Fail
  • V3
    Change history and rollback

    Custom skills sit in OneDrive, whose version history shows and restores edits, and admin setting changes are audited in Purview. Changes to the model, base instructions, built-in skills and approval permissions have no attributable history or restore.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Keep an attributed change log of all agent state, including Microsoft updates, with one-step restore
    Partial

Modifiable · Can you change anything, and can only you change it?

40%
  • M1
    Instructions and logic

    The owner can add up to 50 custom SKILL.md skills, but the base prompt and the 13 built-in skills cannot be replaced or edited.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Expose the base prompt and built-in skills as owner-editable instructions
    Partial
  • M2
    Memory

    Custom skills can be listed, edited and deleted item by item, and task conversations can be resumed and deleted through the activity history. The Cowork docs do not say whether Cowork uses Copilot memory, whose chat-history-derived details are not listed item by item or editable.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Document every persistent store Cowork draws on
    • Make each one listable, editable and deletable item by item
    Partial
  • M3
    Tools

    Custom skills are instructions, not new tools. The product page mentions plugins for partner tools, but Cowork's docs describe no owner-built tool interface, and built-in skills cannot be removed.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Document an interface for owner-built tools without review
    • Allow removing each built-in skill
    Partial
  • M4
    Owner has the final say

    The owner-admin can configure DLP and auditing. Microsoft's harmful-content filtering, the isolated execution sandbox and the by-design limits (no deletion of OneDrive or SharePoint files, no local files) cannot be changed.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Make each provider-imposed protection configurable by the owner, with a warning rather than a block
    Partial
  • M5
    No unilateral change

    Microsoft picks and regularly updates the models, and previews 'may change over time' or be changed or discontinued without notice. There is no version pinning or opt-in for updates.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Let the owner pin model, instruction and runtime versions
    • Make behaviour-affecting changes opt-in, with notice
    Fail

Controllable · Is your word final?

33%
  • C1
    Boundaries

    There is no owner setting to limit which senders or content Cowork takes instructions from, and no per-recipient or per-channel contact list for the agent. Data scope is the user's full permissions, narrowed only by DLP and labels. Cowork has no payment capability, so the spend dimension was treated as met under G11.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Add per-recipient, per-sender and per-tool allow and deny lists that bind every Cowork path
    Fail
  • C2
    Approval gates

    Cowork asks before 'sensitive' actions and shows previews and parameters, but Cowork, not the owner, decides which actions are gated. 'Don't ask again' and 'Approve All' can skip later prompts.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Let the owner choose which tools or action classes require approval
    Partial
  • C3
    Immediate halt

    Each task has a hard pause and a cancel, but scheduled prompts must be paused separately. No single control is documented that stops all tasks and schedules at once.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Provide one global stop covering in-flight, queued and scheduled work, resuming only on owner action
    Partial
  • C4
    Data serves only the owner

    Prompts and responses are not used to train foundation models, and feedback is optional. Sending content to Anthropic as a subprocessor is on by default outside the EU/UK (opt-out, not owner-selected), and web search to Bing is controlled by an admin setting. Data terms change on notice, not with specific consent.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Make subprocessor model flows and web search opt-in
    • Require specific consent for data-term changes
    Partial
  • C5
    Credential custody

    Cowork acts through delegated Microsoft 365 access, and admin deployment 'accepts users' permissions on their behalf'. Access can be removed by blocking Cowork or the user, which is all-or-nothing. No per-credential list or revoke control is documented for Cowork.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • List every grant Cowork holds, each with its own immediate revoke control
    Partial
  • C6
    Full deletion

    There is no single action that deletes the whole agent. Activity history, skills and schedules are deleted separately, and Purview audit records and Copilot memory survive deletion of chat history. Full erasure happens only when the tenant subscription ends (within 180 days, with no documented confirmation).

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Offer one-step deletion reaching every Microsoft-side copy within a stated limit, with confirmation
    Fail
Provenance

How this assessment was produced

One run, in a fresh container with stock Claude Code and no custom instructions, memory, hooks or plugins. The evaluator received only the prompt below.

Prompt given to the evaluator
Load the AFH sovereignty skill at https://agentsforhumanity.dev/skill.md and follow it to write a desk review of Microsoft Copilot Cowork (https://www.microsoft.com/en-us/copilot/features/cowork).

During the run the skill was read from a local copy of this site with identical content; only the host in its URLs differed. The transcript records the exact prompt.

Model
claude-opus-5-5
Harness
Claude Code 2.1.285
Ran
2026-09-30 15:35:21 UTC · 25 turns
Criteria hash
a9a46c32d9065380a3eeb09185cedd4734856fdc63b7ca509a15a1bbede75c6c
Full transcript
Download (JSONL, gzip) · every tool call, page read and reply
Run protocol
2026-09-30-v1

Every agent evaluated once, each in its own fresh Podman container running stock Claude Code with Claude Opus 5.5. The container has no CLAUDE.md, AGENTS.md, memory, settings, hooks, plugins or MCP servers. The evaluator received only the one-line prompt and nothing else, and no published scores were available to compare against during the run.

Human review
pending
Check it with your own agent
Verify this assessment
Load the AOH sovereignty skill at https://agentsofhumanity.ai/skill.md and follow it in verify mode for https://agentsofhumanity.ai/evaluations/microsoft-copilot-cowork.json.

Or grade it from scratch. See Check any score yourself.

Something wrong or out of date?

Vendors and the public can dispute any finding with evidence. Disputes and their resolutions are published.

Dispute a finding