Enterprise AI assistant platform report, November 2026: the bill moved from the seat to the meter
If you own Microsoft 365 Copilot, Claude Enterprise, Gemini for Workspace or ChatGPT Enterprise for a company, your monthly report to leadership almost certainly counts seats: purchased, assigned, active, reclaimed. In the six weeks covered here, three of the four vendors moved the bill off the seat and onto the meter, and all four handed administrators a way to deploy Model Context Protocol connectors into the assistant. A seat count no longer tells you what you are spending or what the assistant can reach. That is the claim of this first issue, and the forward calendar below is where it bites.
Why now? Because the thing the vendors are selling this quarter is not a chat box but an agent that runs tools, reads connectors and executes code on your behalf, and the compute behind that is unpredictable per user. A flat seat price cannot carry it, so the meter arrives with the agent: Copilot Credits with Cowork, the single Enterprise seat plus usage with Claude's unified experience, pay-as-you-go above quota with Gemini Enterprise's third-party models. The tradeoff lands on the platform owner. A cap that is too loose produces an invoice nobody approved; a cap that bites mid-task produces a support ticket and a user who stops trusting the tool. The only way through is to set the policy before the default does.
What this issue is built on, measured rather than asserted: the lane's first live run on 2026-11-01 at 17:10 UTC sent 23 queries, got results from every rail, captured 115 fresh items and dropped none; five parallel researchers read the vendor pages the same afternoon and seeded 85 more; the register stands at 200 rows. The same machinery runs the one-operator fleet that this blog reports on, and the connector questions below are the ones BlindOracle's marketplace asks before an agent is allowed to touch money. We lost something too: the first machine-written narrative mis-stated two dates, which is recorded at the bottom.
What this series will not become. It is not a vendor scorecard and it will not recommend one platform over another; the four are different products with different admin models and the register shows that plainly. There are no affiliate links. It will not publish tenant benchmarks it cannot see, and it will not quote a single adoption statistic as if it had a denominator (the Wharton and Gartner figures appear with their sample sizes, or not at all).
Six decisions the month put on the table
Default-on data egress on three platforms
Gemini's MCP connectors (Asana, Atlassian, HubSpot, Salesforce, QuickBooks…) and the Gemini-app 'Customize' connected apps are ON by default; ChatGPT for Word turned on by default Oct 1; Claude's unified chat+Cowork (cloud sandbox execution, remote control) auto-enables 2026-11-16. Each needs an explicit allow/deny decision per OU or role before users discover it.
Billing moves from flat seats to consumption on 3 of 4 platforms
M365 Copilot Credits default-on Dec 1 (4,000 credits/user/mo default cap) plus Work IQ usage billing; Claude Enterprise becomes one seat + metered usage at the next renewal (20/50-seat minimums, org + per-user caps); Gemini Enterprise pay-as-you-go above quota with per-tier controls and single-user quota multipliers. Spending policies, caps and alert owners must exist before each cutover or the first invoice is the control.
Five retirements with hard dates inside 60 days
Custom GPTs (Dec 11), Claude Sonnet 4.5 + Opus 4.8 fast mode (Nov 30), GPT-5.5 (done Oct 14), ChatGPT sync connectors (done Oct 16), M365 Targeted release (freeze Nov, gone Jan). Each breaks something a user or an automation pins to.
Every platform now ships an admin-deployable MCP/plug-in surface
M365 admin-center MCP connectors, Gemini Agent Gateway custom MCP (VPC/static egress), ChatGPT plugin RBAC with per-role Disabled, Claude Code server-managed MCP allow/deny. One connector-review standard (purpose, data sources, permissions, identity, logging, rollback) should govern all four rather than four vendor-shaped checklists.
Usage-reporting pipelines are changing underneath the monthly report
Copilot Analytics adds feature-level metrics and a Consumption (credits) query; Claude's new Usage & Cost Analytics API makes the old endpoints legacy; ChatGPT Admin Console analytics were restructured Oct 23 (Usage / Insights, new exports); Gemini reports unchanged. Any dashboard built on the old shapes needs re-pointing before the next monthly cycle.
Regulated-tenant exclusions are a planning constraint, not a footnote
Claude's unified experience, Smart reports and the unified seat model each exclude HIPAA-ready / CMEK / ZDR orgs; Gemini Enterprise's 3.8 Flash default routes in-country regions to global unless declined; ChatGPT Sites residency covers US/EU only. Tenants with residency or HIPAA commitments inherit different defaults.
Forward calendar: dated changes and the action each one implies
| Date | Status | Platform | Change | Admin action |
|---|---|---|---|---|
| 2026-10-01 | done | ChatGPT | ChatGPT for Word enabled by default, token-billed at API rates | Verify workspace setting; confirm M365 add-in policy; expect credit draw |
| 2026-10-14 | done | ChatGPT | GPT-5.5 retired from ChatGPT (all plans) | Update role model settings and any SOP naming GPT-5.5 |
| 2026-10-16 | done | ChatGPT | Admin-managed sync connectors retired; real-time Drive/SharePoint/Teams apps replace them | Re-scope real-time apps; set sensitivity-label exclusions; re-test company-knowledge answers |
| 2026-10-26 | done | ChatGPT | New custom GPT creation ended | Inventory GPTs via Usage > GPTs; plan plugin migration |
| 2026-10-27 | done | M365 Copilot | Custom MCP connectors deployable from the admin center; Work IQ APIs move to usage billing | Define MCP authoring authority + review standard; check spending policy auto-apply |
| 2026-11-02 | due | Gemini | Meet real-time translation (63 languages) ON by default at Scheduled Release | Disable for regulated OUs if required |
| 2026-11-16 | due | Claude | Unified chat + Cowork auto-enables for Enterprise orgs unless an Owner decides | Decide: enable early / scope by role / disable; review Remote control, Skip-all-approvals, Scheduled tasks |
| 2026-11-17 | due | Gemini | Gems move to Settings (Skills replace them); Gems stop ≥ 2027-03-01 | Inventory shared Gems + Studio 'Ask a Gem' flows; user comms |
| 2026-11 (day TBD) | due | M365 Copilot | Targeted release enrollment freezes; retired Jan 2027 (Frontier / Standard / Deferred replace it) | Inventory Targeted-release users; pick ring; update runbooks |
| 2026-11-30 | due | Claude | Claude Sonnet 4.5 retired on the API; Opus 4.8 fast mode removed | Inventory pinned model IDs; migrate to 5.5; update role allowlists |
| 2026-12-01 | due | M365 Copilot | Usage-based Copilot Credits default-on for new CSP Copilot Business (4,000 credits/user/mo); standalone SKUs launch, add-on SKUs end-of-sale | Pre-create spending policy + caps + alert recipients; re-key license automation to SKU IDs |
| 2026-12-01 | due | Gemini | Slides whole-deck editing promotional limits end | Budget / comms; evaluate AI Expanded Access add-on |
| 2026-12-11 | due | ChatGPT | Custom GPTs retired (approved deferrals: 2027-02-11) | Migrate to plugins; confirm deferral status in workspace notice |
| 2027-01-31 | due | Gemini | Gemini Code Assist annual renewals end (Antigravity replaces) | Decide migration vs sales exception |
| 2027-03-01 | due | Gemini | Gems stop working for business/enterprise | Confirm auto-migration to draft skills carries sharing |
What each vendor changed
First-party items only, grouped by platform. Dates are the vendor's own; where the vendor named a future effective date it is shown as such.
Microsoft 365 Copilot
| Date | Category | Change | Source |
|---|---|---|---|
| 2026-10-27 | admin | Copilot settings surfaced in classic Outlook for Windows (File > Copilot settings) Users can view/adjust Copilot preferences inside classic Outlook; the pane respects existing policies. | learn.microsoft.com |
| 2026-10-06 | connector-agent | Query string and user-identity-mapping rules for Copilot connectors are now self-service editable Admins can change a connector's query string and identity mapping after creation under Settings > Search & intelligence > Data sources instead of opening a support ticket. | learn.microsoft.com |
| 2026-10-27 | connector-agent | Admins can create and deploy custom MCP connectors from the Microsoft 365 admin center Admins can build custom Model Context Protocol connectors in the admin center (Copilot extensibility section) and deploy them org-wide so Copilot can reach proprietary data live via MCP (Roadmap 501585). | learn.microsoft.com |
| 2026-10-27 | connector-agent | Researcher gains Canva connector; connector results get rich entity cards in Search Researcher can now pull Canva content through the connector when enabled; connector hits in Copilot Search render as metadata-rich entity cards. | learn.microsoft.com |
| 2026-11-01 | connector-agent | Declarative agent manifest v1.9 Adds top-level `agent_skills` (Agent Skills standard skill directories), a `Fabric` capability (query/visualize Fabric data in chat), and makes root `instructions` optional. | learn.microsoft.com |
| 2026-11-01 | connector-agent | Cowork adds Grok 4.7 and GPT 6.1 Sol models, finance/accounting plugins, Power Platform + Copilot connectors, non-Microsoft event triggers Model selector gains Grok 4.7 (xAI as independent processor) and GPT 6.1 Sol; Cowork can now use Power Platform and Copilot connectors and start tasks from events in non-Microsoft systems. | learn.microsoft.com |
| 2026-09-23 | licensing | Microsoft ships licence-utilisation tooling: License Requests page, day-level Copilot usage snapshot, "Inactive Copilot users" metric Release notes (updates Aug 26 – Sep 22) add a dedicated Copilot License Requests page in M365 admin center, a daily-refresh Copilot adoption Power BI report, and Viva Insights metrics for inactive Copilot users (no action in 28 days) plus Organizational Messag | learn.microsoft.com |
| 2026-10-01 | licensing | Usage-based billing becomes default-on for new M365 Copilot Business (CSP) — effective date moved to 2026-12-01 New Copilot Business subscriptions bought via CSP will have pay-as-you-go Copilot Credits enabled by default (was Nov 2, now Dec 1) with a default spending limit of 4,000 credits/user/month, covering Cowork, Work IQ APIs and GitHub Copilot Harness; not initial | learn.microsoft.com |
| 2026-10-14 | licensing | Copilot SKUs decoupled from base-subscription prerequisite; existing offers renamed "(add-on)" From Oct 16 current offers are renamed "Microsoft 365 Copilot (add-on)" / "Copilot Business (add-on)"; from Dec 1 new SKUs (own product/SKU IDs, same price) can be bought without a qualifying M365 base subscription, and the add-on SKUs reach end-of-sale in CSP | learn.microsoft.com |
| 2026-10-27 | licensing | Work IQ APIs move to usage-based billing (no pre-assigned license needed) Developers can call Work IQ agents/models per request with fixed or variable credit costs instead of requiring assigned Copilot licenses; a companion change exposes first-party and tenant-defined declarative agents via the Work IQ endpoint. | learn.microsoft.com |
| 2026-10-28 | licensing | Microsoft FY27 Q1: no new seat count; growth now reported as usage Microsoft gave no paid-seat update (last: ">30 million" in the Oct 21 shareholder letter); Copilot customers billing by consumption grew >75% QoQ, daily Copilot users ~+50% QoQ, E7 "millions of paid seats", paid M365 commercial seats +5% with weaker E3 sales e | www.microsoft.com |
| 2026-10-29 | licensing | Partner Center October 2026: Copilot decoupled from base licence; usage-based billing default-on From 2026-12-01 new M365 Copilot and Copilot Business offers can be bought without a qualifying M365 base subscription at the same price (add-on SKUs reach end of sale in CSP; renewals continue); from 2026-12-01 usage-based billing is enabled by default on new | learn.microsoft.com |
| 2026-11-01 | licensing | Cost Management API (Graph `/copilot/costManagement`, beta) Admins can now create/read/update/delete spending policies, scope them to groups, set per-service settings and notify-only threshold alerts, and read per-user credit balances programmatically; requires `CopilotCostManagement-*` permissions and Billing/Global a | learn.microsoft.com |
| 2026-10-06 | reporting | Copilot Analytics: Cowork adoption and impact metrics + Consumption query Copilot dashboard and Consumption dashboard (Microsoft 365 services tab) add Cowork adoption/impact metrics with export; analysts get a new Consumption query (credit usage) and Cowork fields in the Person query. | learn.microsoft.com |
| 2026-10-27 | reporting | New Copilot metrics in Copilot Dashboard / Advanced Analysis Adds actions in the Copilot app, Edge and OneNote plus feature-level use cases (suggested reply, translate, coach, clean data) across Outlook/Word/Excel/PowerPoint. | learn.microsoft.com |
| 2026-11-01 | reporting | Graph usage report `getMicrosoft365CopilotUsageUserDetailByDate` New admin-settings report function returns per-user Copilot usage for a single date (user-date grain) alongside the existing period reports. | learn.microsoft.com |
| 2026-10-22 | retirement | AI Builder capacity add-ons retire from sale From Nov 1, 2026 AI Builder capacity add-ons are no longer sold; existing credits usable to term end; Copilot Credits are the replacement. | learn.microsoft.com |
| 2026-11-01 | retirement | Targeted release retirement — enrollment freezes November 2026, retired January 2027 In November 2026 admins can no longer add/remove/modify Targeted release users; features move to Frontier / Standard / Deferred release managed from the new unified Release Preferences under Copilot > Settings; GCC/GCC High/DoD keep Standard only. | learn.microsoft.com |
| 2026-09-08 | security | CVE-2026-80098 Copilot Studio elevation of privilege (CWE-347 signature verification) MSRC-published cloud-service CVE, CVSS 3.1 9.3 (Microsoft; Tenable lists 10.0 and 2026-09-03), unauthenticated network attacker, "fully mitigated by Microsoft, no customer action". | msrc.microsoft.com |
| 2026-10-06 | security | SharePoint Advanced Management: item-level report for "Everyone" / "Everyone except external users" permissions New Data Access Governance report lists every item shared with the two special groups (the main Copilot oversharing vector) and can be exported. | learn.microsoft.com |
| 2026-11-01 | security | Purview retention for Copilot memory; Purview DLP extended to Cowork (public preview) Per search summary of the October "What's New in Microsoft Copilot" post: admins can configure retention for Copilot memory in Purview Data Lifecycle Management, and DLP policies for the "Microsoft 365 Copilot and Copilot Chat" location now cover Cowork (block | techcommunity.microsoft.com |
Anthropic Claude Enterprise / Team
| Date | Category | Change | Source |
|---|---|---|---|
| 2026-10-23 | admin | Claude Tag (Claude in Slack) API (beta, Enterprise) New `/v1/organizations/claude_tag/` Admin API endpoints read and change Claude Tag settings per Slack channel/workspace; requires an Admin key with `read:claude_tag` / `write:claude_tag` scopes. | platform.claude.com |
| effective 2026-11-16 | admin | "Chat and Cowork unified" experience auto-enables for Enterprise orgs The unified chat+Cowork experience (cloud sandbox code/file execution in any conversation) is off by default today but turns ON automatically on 2026-11-16 unless an Owner/Primary Owner makes an explicit choice; new controls: Scheduled tasks, Remote control, S | support.claude.com |
| 2026-09-15 | connector-agent | Salesforce in Claude plugin (beta) + developer portal for directory plugins Salesforce plugin brings accounts/opportunities/pipeline plus 37 pre-built sales skills into Claude for all paid plans (Salesforce beta sign-up required); developers can submit plugins to the Claude directory via a new portal with review status and usage analy | support.claude.com |
| 2026-09-16 | connector-agent | Claude Design / Slides / Docs in any conversation (beta on Enterprise); Cowork in every conversation (Pro/Max) Users can request designs, decks and documents in any chat including Claude Code and the Artifacts tab; Enterprise gets these in beta; Cowork capabilities merged into standard chat for Pro/Max (Enterprise path is the Nov 16 unified rollout above). | support.claude.com |
| 2026-10-21 | connector-agent | API version 2026-10-21 + major SDK releases (Python 2.0, TypeScript 1.0, Go 2.0; Apache 2.0) New API version defaulted by new SDK/CLI majors with breaking changes, including removal of `temperature`/`top_p`/`top_k` (400 error on 2026-10-21); Admin API methods moved out of beta to `client.organization` / `ant organization` (Sep 30). | platform.claude.com |
| 2026-11-01 | connector-agent | Claude Code server-managed settings + Console→Enterprise migration guidance Owners can push server-managed Claude Code settings from Organization settings > Claude Code (highest precedence, polled hourly), including permissions, MCP allow/deny, `forceLoginOrgUUID`, `allowedProviders` (v2.1.285+), model/effort caps, plugin marketplace | support.claude.com |
| 2026-09-01 | licensing | Enterprise moves to usage-based billing (single Enterprise seat + usage at API rates) Legacy Standard/Premium and Chat / Chat+Claude Code seat types are no longer sold and cannot continue past the next contract renewal; the new single Enterprise seat (includes Claude Code) covers access only, all Claude/Claude Code/Cowork usage is metered at AP | support.claude.com |
| 2026-10-07 | licensing | Monthly API credits for Team (and Max/Ultra) plans Team plans now include a pooled monthly API credit ($20/Standard seat, $100/Premium seat, capped at $500) claimed by linking ONE Claude Console org from Organization settings > Billing; credits expire monthly, cover API/Managed Agents/Agent SDK but NOT Claude | support.claude.com |
| 2026-10-22 | licensing | Claude Models on Google Cloud (Anthropic-managed) The Claude API operated by Anthropic can now be accessed through a Google Cloud project with IAM access control and Google Marketplace billing (Opus 5.5 fast mode included). | platform.claude.com |
| 2026-09-10 | reporting | Smart reports (beta, Enterprise) AI-generated team usage reports (workstreams, cost, friction, repeat patterns worth packaging as skills) over up to 28 days; off by default, enabled under Organization settings > Capabilities > Analytics; optional user-attribution toggle with reveal-name event | support.claude.com |
| 2026-10-29 | reporting | Usage and Cost Analytics API (beta); old Usage & Cost API and Enterprise usage/cost reports marked legacy New `/analytics/usage` and `/analytics/cost` endpoints report tokens and cost by model, workspace, API key and user for Console and Enterprise orgs; prior endpoints remain supported but are legacy with migration guides. | platform.claude.com |
| 2026-09-22 | retirement | Claude 5.5 family available: Opus 5.5 (Sep 22, $4/$20 per MTok), Sonnet 5.5 (Sep 28), Haiku 5.5 (Oct 7) New model line with 1M context, 128k output, adaptive thinking; Opus 5.5 fast mode GA Oct 15 (default rate limits on Start/Build/Scale tiers; also on Microsoft Foundry and Google Cloud); Sonnet 5.5 cache-read price cut to $0.10/MTok (Oct 7); Haiku 5.5 breaks H | platform.claude.com |
| 2026-09-30 | retirement | Claude Sonnet 4.5 retires on the API 2026-11-30; Opus 4.8 fast mode removed 2026-11-30 `claude-sonnet-4-5-20250929` was deprecated Sep 30 with API retirement Nov 30, 2026 (migrate to Sonnet 5.5, which has breaking changes: `between_tools` replaces `disabled` thinking, forced tool_choice errors, account-bound thinking blocks); Opus 4.8 fast mode | platform.claude.com |
| 2026-09-18 | security | Compliance API expansions (Enterprise): Chrome transcripts, Office local sessions GA, unified-Claude chats, Claude Docs as Word, user preferences Compliance API now returns Claude-in-Chrome transcripts (beta, Sep 18), Microsoft 365 Office local-session content GA (Sep 24), unified Claude chats (Oct 8), Claude Docs exports as .docx (Oct 8) and members' "Instructions for Claude" / Cowork "Global instructi | platform.claude.com |
| 2026-10-20 | security | Web fetch admin toggle + org domain allow/block lists; Managed Agents inherit org restrictions Admins can turn web fetch off or set allowed/blocked domains on the Capabilities page; Managed Agents web_search and web_fetch now must pass BOTH org-level and agent-level domain lists; web_fetch limited to URLs already in session context (Oct 7). | platform.claude.com |
| 2026-10-20 | security | Inference hooks GA (Enterprise) + User profiles + enforcement fields + org webhooks (beta) Server-side allow/deny inference hooks on governed prompts are out of beta for Enterprise; new `/v1/user_profiles` lets each API request be tagged to an end customer so Anthropic enforcement targets one customer not the org, with `enforcement_status`/`enforcem | platform.claude.com |
Google Gemini for Workspace / Gemini Enterprise
| Date | Category | Change | Source |
|---|---|---|---|
| 2026-09-22 | admin | Meet "Take notes for me" new admin + user default option Third option auto-takes notes only for meetings with 3+ people; defaults ON for Business Std/Plus and OFF for Enterprise Std/Plus, Frontline Plus, AI Pro for Education; Quick notes one-page summary becomes the default tab. | workspaceupdates.googleblog.com |
| 2026-09-23 | admin | Temporary administrator roles in Admin console Super admins can assign admin roles to users, groups or service accounts with automatic expiry (preset or custom); not for the primary admin. | workspaceupdates.googleblog.com |
| 2026-10-19 | admin | Gemini Notebook external sharing controls incl. Trusted Domains (Classroom) Admins get granular external-sharing controls for Gemini Notebook (OFF by default; domain/OU/group, Trusted Domains) alongside teacher insight summaries. | workspaceupdates.googleblog.com |
| 2026-10-19 | admin | Gemini in Sheets: BigQuery natural-language analysis (no admin control) and Fill with Gemini web citations Gemini in Connected Sheets now answers against full BigQuery datasets ON by default with NO admin control; Fill with Gemini adds cell-level web citations and up to 3,000 cells/request (governed by Gemini-in-Sheets access). | workspaceupdates.googleblog.com |
| 2026-10-22 | admin | Meet real-time speech translation expands to 63 languages Gemini-powered translation with auto language detection is ON by default when the Gemini for Meet setting is enabled; toggle per OU/group under Apps > Google Workspace > Google Meet > Gemini settings. | workspaceupdates.googleblog.com |
| 2026-09-15 | connector-agent | MCP connectors to third-party tools in Gemini in Workspace Gemini side panel (Docs/Sheets/Slides, Chat) can call Asana, Atlassian Rovo, HubSpot, Mailchimp, QuickBooks, Monday and Salesforce via MCP; enabled BY DEFAULT for users with GW access, managed at domain/OU/group under Apps > Google Workspace Marketplace apps > | workspaceupdates.googleblog.com |
| 2026-09-21 | connector-agent | GE agent governance: transfer shared-agent ownership, resource-level IAM, VPC/static-IP for custom MCP Admins can transfer ownership of shared agents (schedules/triggers disabled until new owner re-enables); resource-level IAM roles and custom project-level roles for apps/data stores (Sep 28); custom MCP servers reachable via Agent Gateway with Private Service | docs.cloud.google.com |
| 2026-10-08 | connector-agent | Claude Opus 5.5 / Sonnet 5.5 selectable in GE (off by default, PAYG) Third-party Anthropic models are available in GE AI developer tools (Oct 8) and the GE app (Oct 15, enable in Feature Management); OFF by default, global/us/eu only, billed as consumption against the spend cap, PAYG must be on first. | docs.cloud.google.com |
| 2026-10-21 | connector-agent | Gemini app "Customize" hub: skills + 40+ connected apps New Customize menu bundles custom skills and 40+ external connected apps plus Google apps; users self-connect and can audit/disconnect; admins manage connected apps and third-party integrations at Apps > Additional Google services > Gemini; post does not state | workspaceupdates.googleblog.com |
| 2026-10-23 | connector-agent | Gemini Notebook becomes a Connected App in the Gemini app Users can create/edit/query notebooks from Gemini chat (@Gemini Notebook, or auto-invoked); ON by default (OFF by default in the EEA), governed by the combination of Gemini app ON and Gemini Notebook ON per OU/group. | workspaceupdates.googleblog.com |
| 2026-09-17 | licensing | GE Standard/Plus no longer include Gemini Code Assist New or renewed online GE Standard/Plus subscriptions sold from Sep 17 exclude Code Assist (existing subs keep it until term end); Antigravity is the replacement; separately Code Assist Standard/Enterprise stopped new sales Oct 9 (annual renewals end 2027-01-31 | docs.cloud.google.com |
| 2026-10-06 | licensing | GE pay-as-you-go above quota renamed, per-tier controls, single-user quota caps "Overages" renamed "Pay-as-you-go usage above quota" with per-subscription-tier controls (Oct 6), extended to Frontline/EDU/Emerging Market editions needing an invoiced billing account (Oct 7), and admins can now set a per-edition, per-location single-user quo | docs.cloud.google.com |
| 2026-10-20 | licensing | Antigravity Enterprise Gateway GA (PAYG via service account, not seats) Gateway requests bill pay-as-you-go through a central Google Cloud service account and do not draw on Standard/Plus seat quotas; gateway and Google Account sign-in cannot be combined; admin controls for Antigravity Boost (GA) and Teamwork (Preview) added Oct 6 | docs.cloud.google.com |
| 2026-10-30 | licensing | Gemini in Slides whole-deck editing: promo limits end Dec 1 Multi-slide/whole-deck editing ships with promotional higher limits through at least 2026-12-01, after which per-user usage limits apply (details to come in Help Center); no admin control. | workspaceupdates.googleblog.com |
| 2026-11-01 | licensing | Google Workspace 50% promotional pricing Nov 15 2026 – Feb 15 2027 workspace.google.com/pricing shows Starter $3.50 / Standard $7 / Plus $11 (50% off list $7 / $14 / $22) for the promo window, "AI Expanded Access" add-on with no published price, Enterprise quote-only; Gemini is bundled (Standard: Gmail/Docs/Meet; Plus/Enterpr | workspace.google.com |
| 2026-11-01 | reporting | Gemini reports in Admin console (org/user-level, 28-day, feature + threshold) and Gemini Notebook audit logs No new reporting posts found in the window; current state: Generative AI > Gemini reports gives org- and user-level usage, per-feature adoption and "hit limit" thresholds (Feb 2026), Meet (Jan) and Chat (Aug) metrics, 2–3 day lag; Gemini Notebook audit logs ro | knowledge.workspace.google.com |
| 2026-09-30 | retirement | Skills replace Gems; Gems retirement timeline published Skills (reusable SKILL.md-based instruction sets) are rolling out in the Gemini app (from Oct 13) and Workspace/Studio (Oct 5 Rapid / Oct 19 Scheduled, but marked "currently delayed" on Oct 13); Gems move to Settings on 2026-11-17, stop working no sooner than | workspaceupdates.googleblog.com |
| 2026-09-23 | security | GE model-residency changes: 3.8 Flash default routes in-country regions to global; Gemini 2.5 Pro removed in Canada/Japan Gemini 3.8 Flash is ON by default in global/us/eu and in-country regions must accept a warning that traffic routes to global (no regional residency); Gemini 2.5 Pro got an admin toggle (Oct 1) and was discontinued in Canada/Japan on Oct 20 (sessions fall back | docs.cloud.google.com |
| 2026-10-22 | security | Google Vault filtered exports for Gemini in Workspace conversations Vault can now export Gemini-in-Workspace conversations (Gmail, Drive, Docs, Sheets, Slides, Meet) as JSON, filtered by custodian/OU (≤5,000 users) and date range; only conversations created with conversation history ON are captured; Vault API support follows w | workspaceupdates.googleblog.com |
OpenAI ChatGPT Enterprise / Business
| Date | Category | Change | Source |
|---|---|---|---|
| 2026-09-11 | admin | Group Managers (delegated admin) + Groups Admin API CRUD + Models > Test Owners/admins can designate Group Managers with delegated analytics/usage-limit/permission scope; Groups Admin API now creates/updates/deletes groups; **Models > Test** inspects a member's effective model access. | help.openai.com |
| 2026-09-17 | admin | Tenant-wide SCIM now covers the API Platform Global admins can assign IdP-synchronized groups to API organizations/projects via Admin Console > Product access; members get access without accepting invites. | help.openai.com |
| 2026-09-17 | connector-agent | ChatGPT for Word — **enabled by default from Oct 1**, token-billed Word add-in available to Enterprise/Edu/Business; admin can toggle in admin console, M365 admin must also allow the add-in; usage is token-priced at API rates and draws from the shared Codex/premium allowance. | help.openai.com |
| 2026-09-29 | connector-agent | Teams, Team tasks, admin-configured shared app connections, @ChatGPT in Slack/Teams, Sites automations Admins can configure workspace app connections that a team owner enables for a whole team (no per-user connect); team tasks run on schedule/trigger with approved tools; @ChatGPT bot for Slack/MS Teams uses admin-configured connections; Sites can use connected | help.openai.com |
| 2026-10-01 | connector-agent | Plugins & marketplaces moved to Admin Console; plugin access **by role** (Oct 8); bulk-manage 25 plugins (Oct 15) Owners/admins manage plugins and plugin marketplaces in Admin Console (apps stay under Workspace settings > Apps as "legacy"); RBAC workspaces can set Available / Install / **Disabled** per plugin and per custom role; bulk edit up to 25 plugins. | help.openai.com |
| 2026-10-16 | connector-agent | Administrator-managed sync connectors **retired**; new app content controls Admin-managed sync connectors are gone — use real-time Google Drive, SharePoint, Teams apps; new content controls (incl. Outlook sensitivity-label exclusions) live under Safety & Security Controls per app. | help.openai.com |
| 2026-09-08 | licensing | Business partner-plugin $50 credit promos; Premium seat pricing context Business users can earn a $50 workspace credit (max 2/workspace) by completing a workflow in Stripe/QuickBooks/HubSpot/Gusto/Dropbox/Canva plugins — admin must enable the plugin; Premium seats (Aug 24) are $100/user/mo annual or $125 monthly vs Standard $20/$2 | help.openai.com |
| 2026-09-29 | licensing | OpenAI Marketplace (beta) — commitment drawdown for partner software Eligible enterprise customers can apply part of their OpenAI commitment toward qualifying partner software; contracted/invoiced by the partner, no self-service checkout. | help.openai.com |
| 2026-11-01 | licensing | OpenAI seat tiers, 200-seat Business cap, Enterprise token-based billing Business: Standard $20 annual / $25 monthly, Premium $100 / $125 (5x usage, no 5-hour limit), 2-seat minimum, hard cap of 200 paid seats per Business workspace since 2026-08-24; Enterprise moving from credits to USD token metering with admin workspace budgets; | help.openai.com |
| 2026-10-23 | reporting | Workspace analytics restructured in Admin Console; 7-day grace for personal→workspace migration Analytics now split into Usage and Insights (User surveys replace Impact tab; Benchmarks under Insights); Usage > Overview adds Workplace health; export adds a by-user report plus GPT and Project reports; verified-domain workspaces give users a 7-day grace per | help.openai.com |
| 2026-09-11 | retirement | Custom GPT retirement → plugins OpenAI is retiring custom GPTs; for affected Enterprise workspaces migration tooling targeted Sep 22 2026, **new GPT creation ends Oct 26 2026**, standard retirement **Dec 11 2026** (approved-deferral workspaces: Feb 11 2027); Business workspaces follow worksp | help.openai.com |
| 2026-09-14 | retirement | GPT-5.5 retired from ChatGPT, Work and Codex GPT-5.5 was retired on Oct 14 2026 on all plans incl. Business/Enterprise/Edu; API unaffected. | help.openai.com |
| 2026-09-29 | retirement | New models: GPT-6.1 Sol (Work/Codex), GPT-6 in Chat, GPT-6.1 Terra GPT-6.1 Sol rolled to Work/Codex (Enterprise/Edu admins must enable); GPT-6 (via GPT-6 Sol setting) rolled into Chat with no new admin preview period; GPT-6.1 Terra rolling out Oct 26 to Business/Enterprise/Edu. | help.openai.com |
| 2026-09-11 | security | Codex Policies & Configurations changes now in audit logs / Compliance API Changes made in the Codex Policies & Configurations UI are written to workspace audit logs, retrievable in UI or via the Compliance API. | help.openai.com |
| 2026-09-21 | security | Privacy Center (Business) Signed-in Business users get a Privacy Center summarizing chat privacy, memory, data use, connected apps and account security with links to settings. | help.openai.com |
| 2026-09-24 | security | New **External access** page in Admin Console Global admins get one place to set org defaults/app-specific permissions for Sign in with ChatGPT, third-party apps' access to members' connected apps, ChatGPT Sites use of connected apps, and Ads account data; new permissions were off by default during previe | help.openai.com |
| 2026-10-15 | security | ChatGPT Sites for Enterprise: admin preview, data residency, Tunnels Sites admin preview lets admins decide who can create/share before member enablement; public publishing off by default; new Sites follow US/EU residency (other regions not covered; existing Sites not migrated); **Tunnels** connect Sites to private-network serv | help.openai.com |
| 2026-10-20 | security | Locked chats Members can PIN/biometric-lock chats out of history and search; retention policies still apply and locked chats **remain in Compliance Platform conversation logs**. | help.openai.com |
Security advisories in the window
No in-window vulnerability disclosure was found for Claude Enterprise, Gemini for Workspace or ChatGPT Enterprise connectors. That is a gap in the sources read, not a clean bill.
| Date | Platforms | Advisory | Admin action | Source |
|---|---|---|---|---|
| 2026-09-08 | M365 Copilot (Copilot Studio) | CVE-2026-80098 Copilot Studio elevation of privilege (CWE-347 signature verification) MSRC-published cloud-service CVE, CVSS 3.1 9.3 (Microsoft; Tenable lists 10.0 and 2026-09-03), unauthenticated network attacker, "fully mitigated by Microsoft, no customer action". | None technically; log it in the Copilot Studio risk register, confirm agent inventory + publishing approvals in Power Platform admin center, and ask Microsoft for the root-cause narrative under your support contract (MSR | msrc.microsoft.com |
| 2026-09-17 | M365 Copilot | CVE-2026-85885 (EoP, command injection, CVSS 9.9 CNA / 8.8 NVD) and CVE-2026-78501 (Business Chat info disclosure, CVSS 7.4) Two more September cloud-service CVEs in M365 Copilot: 85885 lets an authorized attacker elevate privileges over the network; 78501 lets an unauthorized attacker disclose information; both server-side fixed, exploitation "No". | No patch to deploy; review Purview audit logs for Copilot activity around Sep 2026 if you have a regulatory duty to assess exposure; CVE-2026-78501 details come from trackers only (MSRC page not retrieved) — `unverified` | nvd.nist.gov |
| 2026-09-08 | All (MCP/AI coding assistants; Gemini named) | GTIG AI Threat Tracker "From Prompting to Autonomy" Google reports UNC6780 publishing trojanized MCP servers to PyPI (e.g. `tiktoken_mcp`) and injecting code into official repos, DUSTMAKER malware using prompt injection in hidden project config files and in loader comments to defeat LLM security scanners, and a credential-harvest campaign run by an a | Treat MCP servers and agent config files as software supply chain: private allowlisted registry/mirror for MCP packages, block workspace-level hook/config auto-execution in dev tools, and add "LLM-scanner-evasion prompt" | cloud.google.com |
| 2026-10-05 | All (any MCP-enabled assistant; enterprise dev hosts) | CloudSEK "Caught in 4K: The Gentlemen Files" — MCP used as ransomware C2 A Gentlemen affiliate ("Azazel") registered a reverse-shell handler as a tool inside an AI coding assistant via MCP, drove intrusions through `exec_in_session` calls, and ran internet-wide scanning for exposed MCP ports; 24+ organisations in 6 countries hit. | Enforce loopback-only binding for local MCP servers, treat `exec_in_session`-class tools as privileged with audit trail, alert on MCP `initialize` RPCs from non-allowlisted clients, and scan egress for MCP ports. | www.cloudsek.com |
| 2026-10-27 | All (self-built MCP servers; AWS API MCP) | MCP SDK bearer-token audience bypass: CVE-2026-116684 (Python SDK, CVSS 6.8), CVE-2026-111771 (TypeScript SDK), CVE-2026-115413 (aws-api-mcp-server, CVSS 7.8) Tokens issued by a shared IdP for another service are accepted by MCP servers; Python fix (mcp 1.30.0 / 2.2.0) adds `validate_token_resource`, which is OFF by default; AWS fix is 1.6.0 with audience/client-id config. | Inventory internal MCP servers using SDK bearer auth; upgrade AND set `validate_token_resource=True` + `resource_server_url`; for AWS API MCP upgrade to ≥1.6.0 and configure expected audience. Upgrading alone does not cl | osv.dev |
| 2026-10-29 | Claude (Claude Code VS Code extension, via Cursor marketplace) | Trojanized Claude Code extension 2.1.312/2.1.313 with persistent SSH backdoor Tampered builds under Anthropic's publisher ID were live ~11 h on Cursor's marketplace (02:27–13:35 PT, Oct 29); ≤5,000 users; payload adds `svc-telemetry` SSH key, persistence, host beacon; Microsoft VS Marketplace, Open VSX and npm not affected; Anthropic systems not compromised. | Hunt for `svc-telemetry` in `~/.ssh/authorized_keys` and the persistence artifacts on any Cursor host that auto-updated Oct 29; isolate, rotate credentials; decide whether Cursor marketplace remains an approved extension | www.helpnetsecurity.com |
List prices as published on 2026-11-01
Verify against your own contract. Anything the vendor does not publish is marked unverified.
| Platform | Plan | List price per seat per month | Billing notes | Source |
|---|---|---|---|---|
| Microsoft 365 Copilot | Enterprise add-on (becomes standalone SKU 2026-12-01, same price) | $30.00 paid yearly; $31.50 paid monthly on annual commitment | Requires qualifying M365 base licence until 2026-12-01; E3 $39 / E5 $60 after 2026-07-01 increase (`unverified` — secondary reporting of Dec 4 2025 announcement); Copilot Credits usage billing optional on Enterprise, default-on for new Copilot Business from 20 | www.microsoft.com |
| Microsoft 365 Copilot | Copilot Business (≤300 users) | $21 annual (`unverified` — reseller pages; Microsoft page not retrieved); 15% promo through 2026-12-31 per Partner Center (~$17.85, reported as "$18") | Promo end date conflicts across sources (Sep 30 vs Dec 31 2026); Partner Center says 15% through Dec 31 2026 | learn.microsoft.com |
| Anthropic Claude | Team — Standard seat | $20 billed annually / $25 monthly | 2–150 seats; Premium seat $100 annual / $125 monthly (5x usage); plan includes up to $500/month pooled API credits; card only | claude.com |
| Anthropic Claude | Enterprise | $20/seat/month billed annually + usage at API rates | 20-seat minimum self-serve (50 sales-assisted per third parties — `unverified`); admins set user/org spend limits; SCIM, audit logs, Compliance API | claude.com |
| Google Gemini (Workspace) | Business Starter / Standard / Plus | $7 / $14 / $22 list; promo $3.50 / $7 / $11 from 2026-11-15 to 2027-02-15 | Gemini bundled (Starter: Gmail only; Standard adds Docs/Meet; Plus adds Sheets/Slides/Drive/Chat); annual option "save 16%" shown without per-seat figure; "AI Expanded Access" add-on price not published | workspace.google.com |
| Google Gemini (Workspace) | Enterprise | Not published (contact sales); third-party estimates $23–36 (`unverified`) | Same Gemini feature set as Plus | workspace.google.com |
| OpenAI ChatGPT | Business — Standard seat | $20 billed annually / $25 monthly | 2-seat minimum; Premium seat $100 annual / $125 monthly (5x usage, no 5-hour limit); hard cap 200 paid seats per workspace since 2026-08-24; optional workspace credits for overage | help.openai.com |
| OpenAI ChatGPT | Enterprise | Not published (contact sales); third-party reports ~$60/seat with 150-seat minimum (`unverified`) | Seat fee + usage; credit-based or token-based per contract; token rate card e.g. GPT-6 Sol $2 in / $0.20 cached / $10 out per 1M tokens, GPT-6 Astra $10 / $1 / $50; Fast mode 2x, Ultrafast 6x; data residency 1.1x | help.openai.com |
What I would do Monday, and what I would not
Do: put a spending policy and a per-user cap in front of every consumption switch before its date (Copilot Credits on 2026-12-01, the Claude Enterprise seat model at renewal, Gemini Enterprise pay-as-you-go above quota). Run one connector review standard across all four platforms, with the same eight questions (purpose, data sources, permissions, identity model, data flow, logging, support owner, rollback) whether the connector arrives through the Microsoft 365 admin center, Gemini's Agent Gateway, ChatGPT's plugin roles or Claude Code's managed settings. Those eight are the same questions a verifiable agent passport and a signed delegation answer for an agent, which is why the BlindOracle audit ladder starts with them; a connector is an agent with a narrower job. If you need an outside opinion on an MCP server before enabling it, the injection-resilience and MASSAT audits exist for exactly that, and the trust page explains what their receipts prove and what they do not. Re-point any usage dashboard built on the old Anthropic usage endpoints or the pre-October ChatGPT analytics tabs before the next cycle.
Do not: treat a default as a decision. Gemini's third-party MCP connectors, the Gemini-app connected apps, ChatGPT for Word and Claude's unified experience each arrive on, or turn on, without an administrator choosing them. The absence of a ticket is not consent. And do not read a seat-utilisation figure as a cost figure once the meter is running; the two diverge on the first invoice.
Scope: checked, not checked, what would change the conclusion
Checked. Microsoft Learn release notes (6 and 27 October), Partner Center October announcements, Targeted-release retirement, Copilot extensibility notes; Claude help-center and platform release notes, unified-experience, Enterprise plan and migration articles; Workspace Updates (September and October), gemini.google release notes, Gemini Enterprise release notes; ChatGPT Enterprise, Business and model release notes (OpenAI pages returned HTTP 403 to a direct fetch, so a cached copy dated 2026-10-22 plus fresher excerpts were used); MSRC, NVD, CloudSEK, Google Threat Intelligence, Gartner and Wharton releases for the cross-platform rows.
Not checked. Any tenant. The Microsoft 365 Message Center. Tech Community post bodies (JavaScript-rendered). Copilot Studio October platform builds. OpenAI entries dated 24 October to 1 November other than the GPT-6.1 Terra note. Workspace Updates for 15 to 17 September. Vendor trust centres and blogs.
Would change the conclusion. A Microsoft or OpenAI entry in the unread window that moves a retirement date. The Copilot-memory retention conflict (a Tech Community summary says Purview retention now covers it; the Learn page still says it does not) resolving either way. A tenant-specific configuration that makes a flagged default inapplicable.
Known defect in this issue. The model-written narrative of the internal version of this report said "the register gives no date" for two items whose dates live in a separate effective-date field (the Claude unified experience on 2026-11-16 and the custom GPT retirement on 2026-12-11). The dates are correct in the calendar above; the prompt was fixed the same day so the December issue carries them in the narrative.
For the agent side of the same problem, the for-agents page and the FAQ cover how an agent presents a person-signed grant to a venue, and the verify page lets anyone check a receipt without an account.
Next issue: 1 December 2026, covering 1 to 30 November. If you administer one of these platforms and a change you lived through is missing here, the correction is welcome.