Admin
Google Cloud Agent Identity Auth Manager Centralises Tool Credentials
SIG-ADMIN-018
Authority | Visibility
Google Cloud has made Agent Identity auth manager and its Agent Identity APIs generally available, providing a central credentials vault and authentication broker for outbound tool authentication while replacing the legacy IAM Connectors API for managing auth providers and agent identities. Teams should identify legacy API use, clarify who controls credential configuration, and verify logging as they adopt the new services.
Admin
Cloudflare AI Adds Optional OAuth Scope Selection For Wrangler And MCP Server
SIG-ADMIN-017
Authority | Visibility
Cloudflare now lets users choose optional OAuth scopes when authorising Wrangler or the Cloudflare API MCP server. Organisations can narrow permissions at consent time, but should review scope-selection and reauthorisation practices for developer and MCP workflows.
Data
Google Workspace Consolidates Drive External Sharing Fields In BigQuery Inventory Reporting
SIG-DATA-012
Visibility
Google Workspace Drive Inventory Reporting in BigQuery now consolidates direct permissions, group memberships and public links into clearer external-sharing fields. Organisations should review how existing queries, dashboards and exposure-monitoring processes use the new reporting fields.
Capability
Claude Python SDK v1.0 Requires Python 3.10 And Removes Legacy APIs
SIG-CAP-024
Continuity | Authority
Anthropic’s Python SDK v1.0 requires Python 3.10 or later, moves its HTTP layer from httpx to httpx2, and removes deprecated Text Completions and Messages sampling parameters. Organisations using the official SDK should check runtime versions, removed API calls, and tracing or mocking dependencies before upgrading so migration work does not disrupt Claude API integrations.
Capability
Slack Code Channels Move Ai Coding Work Into Team Channels
SIG-CAP-020
Authority | Visibility
Slack Code channels bring AI coding work, live diffs, previews and some inline approvals into shared Slack spaces. Organisations should review which repositories and projects can be reached, who can see development context, and how in-channel approvals relate to existing repository controls.
Admin
Slack Add To Slack Opens One-Click Agent Deployment
SIG-ADMIN-013
Authority | Visibility
Add to Slack lets users deploy custom agents from partner creation tools directly into Slack channels while inheriting workspace permissions and surfacing agents in the Agent Browser. Organisations should review which agents can enter the workspace, their data access and who owns ongoing review and removal.
Capability
Google Meet Room Display Mode Separates Presenter Controls From Shared Room Display
SIG-CAP-019
Visibility
Google Meet's beta Room Display mode separates shared meeting content from the presenter's private controller view on eligible extended-display setups. Organisations should review expectations for private Ask Gemini use and how users understand what remains visible in the room.
Control
OpenAI Codex Hooks Can Run Asynchronously And Invoke MCP Tools
SIG-CTRL-003
Authority | Visibility
OpenAI Codex hooks can now run commands asynchronously and invoke MCP tools. The change broadens what configured hooks can do inside developer workflows, making authority over hook definitions and visibility into resulting commands and tool calls more important without establishing that permission or approval defaults have changed.
Capability
OpenAI Codex Adds Amazon Bedrock Runtime As A Built-In Provider
SIG-CAP-025
Authority | Visibility
OpenAI Codex now supports Amazon Bedrock Runtime as a built-in provider with AWS profile, region and GPT-5.6 routing support. Teams can make provider selection part of ordinary Codex configuration, creating practical questions about who controls that setting, which AWS profiles and regions are approved, and how provider choice is recorded.
Capability
Google Cloud Dataform Remote MCP Server Reaches GA
SIG-CAP-018
Authority | Visibility
Google Cloud has moved the Dataform remote MCP server to GA, allowing AI agents to manage data transformation workflows. Leaders should confirm which agents can connect, what actions they may perform, and what approval and attribution records are required.
Admin
Google Cloud A2UI and A2A Agent Registration Reaches GA
SIG-ADMIN-011
Authority | Visibility
Google Cloud has moved A2UI and A2A agent registration in Gemini Enterprise to GA, including A2UI v0.9. Leaders should confirm who can register production agents, who owns them, and what authentication, approval and operational records apply.
Admin
Google Admin Console Adds Gemini Admin Assist
SIG-ADMIN-012
Authority | Visibility
Google has added Gemini-powered Admin Assist to the Google Admin Console, including a sidepanel and Search Overviews for eligible Business-edition Super Admins. The change places generated guidance and synthesized Help Center summaries directly inside privileged administrative workflows.
Admin
Google Cloud Adds Monitoring Metrics For Gemini Semantic Governance Policies
SIG-ADMIN-010
Visibility | Authority
Google Cloud added Preview Cloud Monitoring metrics for Gemini Enterprise Agent Platform semantic governance policies, including ALLOW/DENY verdict distribution, latency and token consumption. Leaders can use the telemetry to set monitoring thresholds while keeping policy verdicts distinct from downstream agent outcomes.
Data
Google Meet Brings Gemini Note-Taking Into In-Person Meetings
SIG-DATA-011
Visibility | Continuity | Authority
Google Meet now lets users launch Gemini note-taking for face-to-face meetings from the Meet home screen on web or mobile, producing a summary, action items and full transcript in Google Drive. Organisations may need to review how existing Meet notes settings apply when physical meetings become persistent records.
Admin
Gemini Enterprise Custom Skills Require Admin Enablement And Sharing Approval
SIG-ADMIN-009
Authority | Continuity
Gemini Enterprise now lets end users create, upload and share reusable custom skills. Administrators control enablement, availability and sharing approvals, creating a governance decision over which user-authored instructions become shared working guidance.
Control
Microsoft 365 Copilot ServiceNow Connectors Now Apply ServiceNow Role Permissions
SIG-CTRL-002
Authority | Visibility
Microsoft 365 Copilot's ServiceNow Knowledge and Catalog connectors now evaluate assigned ServiceNow roles alongside user criteria when checking access. Organisations should review role design and ownership because ServiceNow role changes can now alter what connected content users can retrieve through Copilot.
Governance
Claude Compliance API Now Returns Local Session Transcripts
SIG-GOV-006
Visibility | Authority
Claude Enterprise Compliance API beta endpoints now return transcripts from local Cowork and Claude Code sessions running on users’ machines. This brings local AI session records into central compliance visibility and requires organisations to decide who may retrieve and use them.
Pricing
Claude Managed Agents Can Pause When Session Budgets Are Reached
SIG-PRICE-001
Authority | Continuity
Claude Managed Agents can now enforce hard spending caps on sessions using public list prices. When a cap is reached, the session pauses before starting new model requests, creating a direct link between cost controls and operational continuity.
Capability
Claude Managed Agents Can Consult A Second Model Mid-Turn
SIG-CAP-016
Authority | Attribution | Visibility
Claude Managed Agents can now consult a designated secondary model mid-turn for strategic guidance. Teams may therefore need to account for which advisor model was configured and whether its guidance influenced a session when reviewing how an output was produced.
Capability
Claude Managed Agents Can Load GitHub Repository Skills
SIG-CAP-015
Authority | Attribution | Visibility
Claude Managed Agents can now load skills automatically from a mounted GitHub repository's root .claude/skills directory at session start. Repository changes may therefore alter the instructions or capabilities available to later agent sessions. Leaders should review which repositories are mounted, who can change skills, what review controls apply and how agent behaviour is tested after updates.
Governance
Claude Enterprise Can Hold Governed Prompts For Security Review Before Inference
SIG-GOV-005
Authority | Visibility | Continuity
Claude Enterprise inference hooks can hold governed prompts until an organisation's AI security server returns an allow or deny decision before inference. Leaders should review who owns the security server and rules, how denials are recorded, and how failure handling affects continuity when the service is unavailable.
Capability
Notion Meeting Notes Can Trigger Custom Agent Actions
SIG-CAP-014
Authority | Attribution | Visibility
Notion Custom Agents can now run after an AI Meeting Note is finished, including updating project trackers, posting recaps or creating engineering tickets. This shifts some workflows from user-prompted assistance to event-triggered execution. Leaders should review which meetings can activate each agent, what systems and records it can change, whether validation or human review is required, and how triggered actions are logged and attributed.
Capability
Claude Opus 5 May Fall Back To Opus 4.8 For Flagged Cyber Requests
SIG-CAP-013
Authority | Attribution | Visibility
Anthropic’s Claude Opus 5 may automatically fall back to Opus 4.8 when cyber-safety classifiers flag certain security-related requests on Claude.ai, Claude Code or Claude Cowork. API customers can choose whether to enable fallback and can identify which model served the response. Leaders should review where model-specific approval matters, whether Opus 4.8 is an acceptable fallback and whether organisational records capture which model actually answered.
Data
Google Meet Notes May Preserve Presented Content Without Meeting Recording
SIG-DATA-010
Authority | Visibility | Continuity
Google Meet will soon allow screenshots of presented slides, diagrams and charts to be added to AI-generated meeting notes, even when full meeting recording is not enabled. Administrators can configure the setting before the Q3 2026 rollout. Leaders should review whether screenshots are permitted in unrecorded meetings, who receives notes containing captured images, and whether existing access, classification and retention rules cover this new form of visual record.
Admin
GitHub Copilot Linear Issue Assignment Can Initiate Autonomous Code Changes
SIG-ADMIN-008
Authority | Attribution
GitHub Copilot can now begin coding work when a Linear issue is assigned to the agent, using the issue content to create a draft pull request in GitHub. Human review and merge controls still apply, but the initiation point has moved into a project management tool. Organisations should confirm who can trigger these sessions, what guidance shapes them, and whether existing access and approval controls cover work initiated from Linear.
Admin
GitHub Issue Approvals Do Not Enforce Agent Permissions
SIG-ADMIN-007
Authority | Visibility
GitHub Issues now offers approval panels, confidence ratings and automation settings for agent-driven changes, but GitHub states that approvals are a workflow convenience rather than a security control. An agent with permission to change issues may still act directly without submitting the change for review. Leaders may need to confirm which agents can change issues, which actions can occur automatically, whether direct and approved changes are distinguishable in records, and whether existing GitHub agent approvals cover this authority.
Data
Google Calendar Now Surfaces Delegate Identities In Meeting Guest Lists
SIG-DATA-009
Attribution | Visibility
Google Calendar now shows when a meeting guest has a calendar delegate, allowing people viewing the guest list to identify and contact that delegate directly. With no admin or end-user control to disable the feature, leaders may need to review which delegated calendars are used for external or sensitive meetings, whether visible support relationships are appropriate, and whether existing delegation guidance covers this new level of visibility.
Admin
GitHub Copilot Brings Adoption Cohorts Into A Visual Dashboard
SIG-ADMIN-006
Authority | Visibility
GitHub has added a visual dashboard that groups Copilot users into vendor-defined adoption categories, compares activity between groups and recommends actions to deepen adoption. Leaders may need to define whether the data will support licence optimisation, enablement, rollout planning or workforce assessment, who can access and share it, and what organisational context must accompany comparisons before GitHub’s categories become embedded in internal reporting and decisions.
Governance
Google Requires Affected Agent Studio Apps To Be Regenerated And Redeployed
SIG-GOV-004
Visibility | Continuity
Google has fixed a security vulnerability in backend code generated for Agent Studio web apps, but previously deployed copies will not update automatically. Organisations that generated, downloaded or deployed affected code before 1 July 2026 must identify those apps, regenerate and test them, then deploy the updated version. Leaders may need to confirm that deployment records are complete, remediation is owned and security reviews cover code generated by AI platforms.
Admin
GitHub Brings Copilot Credit-Pool Controls Into The Billing UI
SIG-ADMIN-005
Authority | Visibility
GitHub administrators can now manage cost-centre AI credit pools through the billing interface, rather than only through the REST API. Organisations already using API automation may now have manual and automated routes capable of changing the same setting. Leaders should clarify who can make changes, which route is authoritative, how settings will stay aligned, and whether billing administrators have approval to stop Copilot usage or permit additional spend.
Admin
Anthropic Workbench Retirement Creates An Export Deadline
SIG-ADMIN-004
Visibility | Continuity
Anthropic will retire legacy Workbench on 17 August 2026, after which saved prompts, variables, completions and evaluations will no longer be accessible through it. The refreshed Workbench is stateless and cannot import legacy data, while several experimental prompt-tool endpoints will also retire. Leaders should identify and export important assets, decide where they will be stored and governed, confirm replacement workflows preserve required evidence, and check whether existing Anthropic approval covers the changed storage model.
Capability
Notion Agent Can Now Move Meetings And Send Invites
SIG-CAP-012
Authority | Attribution
Notion Agent can now manage connected calendars from chat, including moving meetings, creating scheduling links and sending invitations. While it remains bounded by the user’s calendar rights, the shorter path from instruction to action may make Agent-led scheduling routine. Leaders may need to review which calendars can be connected, how Agent actions appear in records, whether existing Notion AI approvals cover calendar write actions, and whether teams are becoming reliant on it as their default scheduler.
Admin
Google Meet Auto Note-Taking Setting Will Be On By Default For Some Business Plans
SIG-ADMIN-003
Authority | Visibility
Google is adding an admin setting that automatically starts “Take notes for me” in meetings with three or more participants. It will be on by default for Business Standard and Business Plus from 21 September 2026, unless administrators change it. Leaders may need to review which teams should use automatic notes, whether existing approvals still cover the changed default, and how meeting-note access, retention, privacy and staff communication are managed.
Capability
GitHub Copilot Auto Model Selection Reduces Model Visibility
SIG-CAP-011
Attribution | Visibility
GitHub Copilot auto model selection allows enterprise administrators to make automatic model choice the default for new conversations, while Copilot CLI can route tasks to different models based on availability, reliability and task type. The issue is not only that developers may use more than one model, but that organisations may lose visibility over which model handled a specific prompt, coding task or agent interaction. Leaders may need to review whether Copilot governance still matches practice, especially where sensitive repositories, model-provider assumptions or fixed-model controls matter.
Data
Copilot Chat in Outlook Reasons Across Inbox and Calendar Without Copilot Licence
SIG-DATA-008
Authority | Visibility
Microsoft 365 Copilot Chat in Outlook can now reason across a user’s inbox, calendar, meetings and accessible Microsoft 365 enterprise data, not only the current email thread. The issue is that organisations may have governed Copilot around licensed users, while affected Outlook users may now access broader AI assistance without a Microsoft 365 Copilot licence. Leaders may need to check who can use the feature, what Outlook context Copilot can draw on, and whether existing policies still match the real user population.
Data
Google Voice AI Note Taking Creates Call Records
SIG-DATA-007
Attribution | Visibility | Continuity
Google Voice AI note-taking can turn calls into transcripts, summaries and action items, with some follow-up content sent through Gmail and stored in the Voice app. The issue is that spoken work may now become written work inside Google Workspace, raising practical questions about what counts as the call record, who can access it, what gets retained and how AI-generated action items are reviewed.
Capability
Claude Tag Slack Memory Initiative
SIG-CAP-010
Authority | Visibility | Continuity
Claude Tag brings memory and initiative into Slack, allowing a channel-level AI participant to remember relevant context, receive delegated tasks and proactively follow up where ambient behaviour is enabled. The issue is not just what Claude can automate, but how leaders can use this shift well: clarifying which work should be handled by AI, which decisions need human judgment, and how teams stay visible, coordinated and accountable as AI becomes part of the channel workflow.
Risk
Anthropic Model Shutdown
SIG-RISK-003
Visibility | Continuity
Anthropic disabled access to Fable 5 and Mythos 5 for all customers after a US export control directive targeted foreign national access. The issue is not only the shutdown itself, but the continuity assumption it breaks: organisations may depend on frontier models that can become unavailable because of government direction, provider compliance decisions or jurisdictional restrictions.
Data
Google Vault Gemini App Retention
SIG-DATA-006
Visibility | Continuity
Google Vault now covers retention rules and litigation holds for the standalone Gemini app, but not embedded Gemini features in Workspace apps such as Help me write in Gmail or Docs. The issue is whether organisations understand which Gemini interactions can be retained, searched and held through Vault, and which remain outside this update.
Data
Claude Fable 5 Data Retention
SIG-DATA-005
Authority | Continuity
Claude Fable 5 requires 30-day data retention on the Claude API and is not available under zero data retention. The issue is whether organisations can use the most capable model while still meeting their data minimisation, client contract or internal retention requirements. For some teams, model selection may no longer be only about capability. It may also depend on how long API inputs and outputs are retained after processing.
Capability
Copilot Studio Computer Use
SIG-CAP-009
Authority | Visibility
Microsoft Copilot Studio now brings computer use to low-code agent builders, allowing agents to operate web and desktop applications through the visible interface. The issue is whether identity, inventory, audit and access controls are keeping pace with who can now build agents that control business applications.
Data
Gemini Drive Gmail Sources
SIG-DATA-003
Authority | Visibility
Gmail threads can now be used as sources in Ask Gemini in Drive, alongside files and folders. The issue is whether existing Gemini in Drive approvals, staff guidance and data handling assumptions included email content as part of the AI reasoning context.
Risk
Anthropic Managed Agents
SIG-RISK-002
Authority | Visibility | Continuity
Anthropic Managed Agents may split tool execution from orchestration, meaning parts of an agent workflow may sit inside customer infrastructure while coordination, context management and session logging remain with the vendor.
Control
Claude Instruction Authority
SIG-CTRL-001
Authority | Visibility
Claude API applications may now change system-level instructions during a conversation, shifting the instruction layer that shapes later responses.
Data
Gemini Drive Sharing
SIG-DATA-004
Authority | Visibility
Gemini app content may now be shared through Google Drive.
Governance
OpenAI Codex Business Workflows
SIG-GOV-003
Authority | Visibility
M365 Copilot may enable Anthropic models by user group.
Capability
Microsoft Scout Always-On Agent
SIG-CAP-008
Visibility | Continuity
Microsoft Scout introduces an always-on personal agent across Microsoft 365 that can build context about how a user works over time. The issue is whether existing data access approvals contemplated ongoing work-pattern learning, not just prompt-based assistance.
Admin
OpenAI Admin API Controls
SIG-ADMIN-002
Authority | Visibility
OpenAI’s Admin API now offers more granular controls for spend alerts, model access, data retention, hosted tool permissions and billing attribution. The issue is whether organisations know these controls exist, who owns their configuration, and how changes are documented.
Governance
M365 Copilot Anthropic User Groups
SIG-GOV-002
Authority | Visibility
M365** **Copilot may enable Anthropic models by user group.
Risk
OpenAI MCP Tunnels Private Infrastructure
SIG-RISK-001
Authority | Visibility
AI agent tunnels may change private system access boundaries.
Capability
OpenAI Codex Locked Machine Operation
SIG-CAP-006
Authority | Visibility | Continuity
Codex may continue tasks after a user locks their Mac.
Capability
OpenAI Codex Can Now Operate Windows Desktop Apps
SIG-CAP-007
Authority | Visibility
Codex computer use now extends to Windows desktop apps, allowing an agent to see, click and type on the active desktop. The issue is not only what Codex can do, but what it may see while operating a work machine.
Data
ChatGPT Spreadsheet Sidebar
SIG-DATA-001
Attribution | Visibility
ChatGPT sidebar in Excel and Sheets may embed AI-Generated logic in spreadsheets organisations treat as human authored.
Data
ChatGPT Connected Context
SIG-DATA-002
Visibility | Continuity
ChatGPT may draw on wider prior context not re-entered by the user.
Capability
ChatGPT Multi-Step Tasks
SIG-CAP-005
Authority | Visibility
GPT-5.5 may now complete multi-step tasks with less user direction than before.
Capability
Claude Agent Memory
SIG-CAP-004
Visibility | Continuity
Managed agents may no longer start from a clean slate.
Capability
Claude, Opus 4.7 Upgrade
SIG-CAP-003
Visibility | Continuity
Model upgrades may change what is visible and what is billable.
Admin
Microsoft Copilot Task Execution
SIG-ADMIN-001
Authority | Attribution
Copilot is enabled to complete defined workflow actions within workplace tools.
Governance
ChatGPT Shared Mailbox Access
SIG-GOV-001
Authority | Visibility
Outlook integration may extend shared mailbox use into a new interface.
Capability
ChatGPT, Context Variability
SIG-CAP-002
Visibility | Continuity
Some responses may not use memory or prior context, creating variability without a visible signal.
Capability
Claude, Persistent Memory
SIG-CAP-001
Visibility | Continuity
AI memory changes assumptions about context, continuity, and visibility.