Skip to content
NewWatch the two minute introduction to EndLayerWatch now →
Changelog

What's new

AI changes every week, and EndLayer keeps pace: the new tools, agents and integrations we cover, and why each one matters for governance.

Updated every weekLatest 20 November 2026

  1. 7 updates · 2 highlights

    Slack, Microsoft Foundry and framework profiles

    Alerts and the weekly briefing reach Slack, Microsoft Foundry joins the cloud connectors, Copilot Studio protection reads tool descriptions, and every framework gets a page an auditor can use.

    • HighlightAddedGovern

      Slack: risk alerts, personal messages and the weekly briefing

      Risk alerts post to a Slack channel you choose, each person gets a direct message about their own devices, and the EndLayer Briefing posts every Monday. Each part is switched on separately, and the app asks for no permission to read messages.

      Why it mattersA policy only works when the right people hear about it, so decisions now reach administrators and employees where they already work.

    • HighlightAddedSurface

      Microsoft Foundry and Azure OpenAI

      Connect your Azure subscriptions to see every Foundry and Azure OpenAI resource, its model deployments, its content filters and who can call each model, with usage and billed cost by model. Each resource is checked for API keys, internet exposure and content filtering, and an activity log records what changed between syncs.

      Why it mattersModels deployed in the cloud join the same inventory and the same cloud policy as the AI on your devices.

      Learn more →
    • AddedGovern

      Copilot Studio: tool descriptions are checked too

      Runtime protection now checks the description of each tool a Copilot Studio agent calls against your data protection policy, as well as its inputs. The reason shown says where the match was found.

      Why it mattersA tool description is text the agent reads on every turn, so a poisoned MCP tool is caught before the agent acts on it.

      Learn more →
    • AddedMonitor

      Tools each agent calls, and a review priority

      Each Copilot Studio agent's profile lists the tools it called in the last 30 days, MCP tools marked, with how many calls were refused or warned. The Agents table adds a review priority score, a column picker and a rows-per-page choice.

      Why it mattersReviewers start with the agents that need them most, with the evidence on one page.

    • AddedGovern

      A profile page for every framework

      Each framework opens its own page: a coverage radar by section, a control-point table for SOC 2, the EU AI Act, ISO/IEC 42001, the NIST AI RMF and GDPR, the policies that would close the gaps, and an export for auditors as a workbook or CSV.

      Why it mattersCoverage against a framework becomes something you can show, section by section, and hand to an auditor.

      Learn more →
    • ImprovedGovern

      Prompt scanning grading you choose

      Administrators choose how a gap in prompt scanning is graded: off, warn or fail. A browser gap names each browser and why, and these warnings sit apart from overall posture.

      Why it mattersPosture reflects real risk, so a rollout still in progress doesn't read as a failing fleet.

    • ImprovedMonitor

      More AI website prompts carry their model

      The Layer 1 Agent and the browser extension recognise newer Claude, Gemini and ChatGPT model names, so more prompts on AI websites carry their model and their cost.

      Why it mattersCost and usage reports are more complete, model by model.

      Learn more →
  2. 6 updates · 1 highlight

    Cloud agents, owned and protected

    Amazon Bedrock AgentCore gains runtime protection, cloud agents gain named owners, and policy editing gets simpler.

    • HighlightAddedGovern

      Amazon Bedrock AgentCore runtime protection

      Tool calls passing through an AgentCore Gateway are checked against your data protection policy, the way Copilot Studio's are. Monitor records what would be stopped; Enforce stops it. Setup is a template you deploy in your own account.

      Why it mattersOne data protection policy now covers coding agents on devices and agents running in the cloud.

      Learn more →
    • AddedSurface

      Cloud agents have owners

      The owner of a Copilot Studio agent or an AgentCore runtime now appears as a person, on Monitoring → Users and on their own page, beside the devices they use.

      Why it mattersEvery agent has someone accountable for it, which is where any review starts.

    • AddedMonitor

      AgentCore activity and drift

      AgentCore has its own page, with each runtime's and gateway's activity, a Drift tab showing what changed in its configuration, and Sync now.

      Why it mattersA change to an agent's configuration is visible the day it happens.

    • AddedMonitor

      An Agents flow on Home

      Home shows the Copilot Studio and AgentCore agents whose tool calls runtime protection caught, with each agent's owner and its logged, warned and blocked totals.

      Why it mattersLeaders see agent risk at a glance, next to device events.

    • ImprovedGovern

      One On/Off button on every policy

      Every policy editor has a single On/Off button. A list policy can be switched off without deleting its entries, and comes back as it was.

      Why it mattersPausing a rule is safe and reversible, so policies can be adjusted with confidence.

    • AddedGovern

      Enable all or disable all data protection rules

      One menu switches every prompt or browser rule on at Monitor, or off, after a confirmation that lists what changes.

      Why it mattersA new policy starts by recording, before it interrupts anyone.

  3. 8 updates · 2 highlights

    Copilot Studio, Teams and kill switches

    Runtime protection for Copilot Studio, alerts in Microsoft Teams, the EndLayer Briefing, kill switches aimed at a group, and new connections for Jamf Pro and Amazon Web Services.

    • HighlightAddedGovern

      Copilot Studio runtime protection

      Every tool call a Copilot Studio agent is about to make is checked against your data protection policy. Monitor, the default, records what would be blocked; Enforce stops it.

      Why it mattersThe same rules that protect prompts on devices now apply to the agents your organisation builds.

      Learn more →
    • HighlightAddedGovern

      Kill switches for a group

      A kill switch can now halt the AI agents on the devices of one Entra group, with a recorded reason, and release restores each device as it was. Members can request a policy change, and administrators answer in the Inbox.

      Why it mattersResponse can be as narrow as the problem, and every change has a reason on record.

    • AddedGovern

      Risk alerts in Microsoft Teams, and the EndLayer Briefing

      Risk events post to a Teams channel, each linking to its review. Every Monday, owners and administrators get the EndLayer Briefing: the week's figures, what needs attention and what's new.

      Why it mattersOversight becomes a weekly habit, with a clear owner for each finding.

    • AddedMonitor

      Anomaly detection

      EndLayer flags the same person, agent or AI website being warned or blocked many times in a short period, and a sudden surge against its own normal level.

      Why it mattersPatterns that need a conversation surface on their own, not only single events.

    • AddedSurface

      Jamf Pro and Amazon Web Services connections

      Jamf Pro mirrors your Macs and shows which ones the agent hasn't reached yet. Amazon Web Services connects through a read-only role and lists your AgentCore runtimes and gateways, graded for authentication, network mode and credentials.

      Why it mattersCoverage gaps and cloud agents show up in the same inventory as everything else.

    • AddedMonitor

      Legal AI: Harvey and Legora profiles

      Harvey and Legora have their own profile pages, with their data protection events and the documents shared with them, and Events can be filtered by website.

      Why it mattersFirms see how legal AI is used, person by person, before setting a rule for it.

      Learn more →
    • AddedGovern

      Uninstall from one machine, and export the policy library

      An administrator can uninstall an AI tool from one machine or from one person's machines. The policy library exports as Excel or CSV, every setting included.

      Why it mattersActions are precise, and the full policy record can leave the console for review.

    • ImprovedSurface

      Scan now on Shadow AI

      Shadow AI shows when the last scan ran, what is new since then, and a Scan now button.

      Why it mattersA new unapproved tool is visible the moment it appears.

      Learn more →
  4. 7 updates · 1 highlight

    A Shadow AI page for undecided AI

    Shadow AI gets its own page, AI websites and MCP servers can be ruled on, and risk events can reach people by email.

    • HighlightAddedSurface

      Shadow AI page

      A new Shadow AI page under Workspace lists the AI nobody has ruled on yet: apps and agents with no decision, MCP servers, AI websites and tools still installed after being turned down. It opens on the Activity flow, with the undecided band highlighted.

      Why it mattersThe AI that has never been reviewed is the first thing to look at, and now it sits in one place.

      Learn more →
    • AddedGovern

      Decisions for AI websites and MCP servers

      An AI website can be marked sanctioned, unsanctioned or blocked from its row or in bulk, and an MCP server can be sanctioned or unsanctioned like an AI tool. An unsanctioned MCP server fails the MCP servers check wherever it is declared.

      Why it mattersEvery kind of AI on the Shadow AI page can now get a recorded decision, not just apps.

      Learn more →
    • AddedSurface

      Microsoft Agent 365 registry

      Connections can read your tenant's Microsoft Agent 365 registry daily: each Copilot agent with who built it, who it is deployed to and its recent use. Agents with no owner, no recent use or deployment to everyone are flagged and listed under Monitoring, Agents.

      Why it mattersAgents built in Microsoft 365 sit in the same inventory as the AI on your devices.

      Learn more →
    • AddedGovern

      Prompt injection in uploaded documents

      A new browser activity template flags a document attached to an AI website that carries instructions written for the model, including hidden text. The document is read on the device, and a warning shows the person where the instruction is and how it was hidden.

      Why it mattersInstructions hidden in a file can steer an AI assistant without the person seeing them.

      Learn more →
    • AddedGovern

      Email notifications for risk events

      Settings, Notifications emails people when EndLayer records a risk event, such as sensitive data blocked or an unapproved AI tool. By default each event goes to the device's primary user, or an administrator can send everything to one address. Notify on an event now sends the email directly.

      Why it mattersThe person who can fix a problem hears about it without anyone copying details across.

    • AddedMonitor

      Claude Code allowances and spending caps

      FinOps shows Claude Code session and weekly allowance pressure, limit hits and reported spending caps beside each person's seat value. A missing reading shows as unknown.

      Why it mattersYou can see who is running into limits before deciding to change a plan.

      Learn more →
    • ImprovedMonitor

      One Apps table per computer and person

      Installed tools and AI websites now sit in one Apps table with a Type column. Downloaded Excel workbooks and PowerPoint decks are also checked for sensitive data on the device.

      Why it mattersEverything a person uses with AI reads as one list.

  5. 7 updates · 2 highlights

    Browser AI usage and outside alerts

    AI websites now count as usage, alerts from Zscaler, Defender, Purview and ServiceNow arrive in Events, and warnings can carry your own words.

    • HighlightAddedMonitor

      AI website prompts count as usage

      Prompts sent on AI websites such as ChatGPT, Claude and Gemini now count as usage, with estimated tokens, the model the site shows and the project or Gem the chat belongs to. Only numbers and those labels leave the page; reply and conversation text never do.

      Why it mattersWork done in the browser is no longer missing from your AI usage and cost picture.

      Learn more →
    • HighlightAddedGovern

      Alerts from Zscaler, Defender, Purview and ServiceNow

      Zscaler DLP, Microsoft Defender, Microsoft Purview and ServiceNow alerts now arrive on Events with a Source column. A ServiceNow connection can also write EndLayer events as records, with the person who owns the machine filled in.

      Why it mattersAI risk sits beside the security signals your teams already work from.

    • AddedGovern

      Your own words on warnings and blocks

      Settings, Notices sets the title and message people see on a warning or a block, with your logo, colour and the popup's position, and a live preview. The same applies in the browser.

      Why it mattersPeople see a message from their own organisation, which makes the rule easier to follow.

    • ImprovedGovern

      Triage in Needs attention

      Issues can be assigned, acknowledged or snoozed, one at a time or in bulk, and you can rate each control's urgency for your organisation. Waiting machines can be approved in bulk, and Inbox, Blocked software lists the blocked apps and MCP servers nobody has decided on.

      Why it mattersA large fleet's issues get an owner and an order instead of one long list.

    • AddedMonitor

      Risk Matrix and AI Adoption

      Insights, Risk Matrix plots machines, people or teams by AI adoption against the issues open on them. AI Adoption gains Highlights and a Teams view, and both have their own reports.

      Why it mattersYou can see which teams use AI most and which carry the most open risk.

    • AddedSurface

      Microsoft SCCM package

      Connections, Microsoft SCCM generates a ready-to-import Configuration Manager package with the signed Layer 1 Agent, install and uninstall scripts, the browser extension policy and a deployment guide.

      Why it mattersOrganisations that deploy with Configuration Manager can roll out the agent with the tool they already use.

    • AddedSurface

      Projects tagged AI or not AI

      Each project is tagged AI or Not AI from the model SDKs, agent frameworks and ML libraries it declares, and project profiles gain an SBOM tab you can download.

      Why it mattersYou can tell which code your organisation is building with AI.

  6. 3 updates · 1 highlight

    Rules for computer use and impersonation

    New policies cover agents that drive browsers or act as the person, and Monitoring gains a view of every project agents work in.

    • HighlightAddedGovern

      Computer use and impersonation policies

      Five new rules, each Off, Warn or Block, cover agents that drive a browser, control the desktop, run in agentic browsers, send messages as the person or use the person's own credentials. The Layer 1 Agent refuses or warns in Claude Code, Codex and Cursor, and each incident appears as an event.

      Why it mattersAgents that act as a person carry that person's access, so these actions need their own rules.

      Learn more →
    • AddedSurface

      Every project AI agents work in

      Every repository the fleet's agents work in is listed in one place, with its files and dependencies. Projects also include the folders people grant to Claude Cowork.

      Why it mattersYou can see what code and folders AI agents are working in across devices.

    • AddedGovern

      Sensitive data checks in Codex CLI

      The sensitive data scanner now runs in OpenAI's Codex CLI as well as Claude Code, Cursor and the browser.

      Why it mattersOne data protection policy now covers another widely used coding agent.

  7. 6 updates · 1 highlight

    Browser activity policies and confidence scores

    Browser activity policies, a confidence score on every sensitive data detection, an audit trail and FinOps by person.

    • HighlightAddedGovern

      AI browser activity policies

      Opt-in templates cover copy, paste, input, uploads, downloads, navigation and clicks on AI websites, each set to Monitor, Warning or Block. Warning asks the person whether to continue, and completed downloads are checked for sensitive data on the device.

      Why it mattersData reaches AI through more than prompts, so the browser rules now cover the other ways too.

      Learn more →
    • AddedGovern

      Confidence scores and a block threshold

      Every sensitive data detection carries a confidence score, shown as a bar in Events. A detector set to Block can carry a minimum confidence, and matches below it warn instead.

      Why it mattersYou can block what is clearly sensitive and warn on what is less certain.

    • AddedGovern

      An audit trail of every admin change

      Every administrative change, from policy publishes and role changes to device approvals and kill switches, is listed in the organisation's audit trail.

      Why it mattersYou have a record of who changed what, ready for review.

    • ImprovedGovern

      One AI tool access policy

      AI tool access is one policy with Sanctioned, Warn and Block choices, keeping existing assignments and exceptions.

      Why it mattersEach AI tool has one clear standing across the fleet.

    • AddedMonitor

      FinOps, People and AI Adoption

      Usage is now FinOps, and FinOps, People shows the bill by person, with the seats each holds. Insights adds rankings of people, apps and agents and a wall of people tiles, and Monitoring, Users is now by person.

      Why it mattersAI cost and adoption read by the people behind them, not only by machine.

      Learn more →
    • AddedSurface

      Harvey, Legora and Z.ai recognised

      The legal AI assistants Harvey and Legora, and Z.ai, join the AI website catalogue and can be picked in the websites browsers must block.

      Why it mattersLegal teams can see and rule on the specialist AI tools they use.

      Learn more →
  8. 3 updates · 1 highlight

    Attack Surface and the real cost of AI

    Attack Surface maps how AI tools, agents and people connect, tool plans show what AI actually costs, and every device shows what AI touches.

    • HighlightAddedSurface

      Attack Surface

      Insights, Attack Surface draws the whole fleet as one graph, read the way an attacker would, with filters for perimeters, kinds and flags.

      Why it mattersIt shows how AI tools, agents and people connect, and where a path could be abused.

    • AddedMonitor

      Plans and real cost per tool

      Set each tool's plan on Apps to see what it actually costs, with a Plan column and usage value, extra usage and seat cost shown side by side. Claude Code and Codex gain the spend dashboard and People view.

      Why it mattersSpend reflects what you pay, not just list prices.

      Learn more →
    • AddedSurface

      A Links tab on every device

      Each device shows its AI apps, agents, MCP servers, skills, repositories, failing controls and sensitive data detections as a searchable graph.

      Why it mattersYou can see everything AI touches on one machine at once.

  9. 3 updates · 1 highlight

    Sensitive data checks in the browser

    Prompts on AI websites can be checked for sensitive data in Chrome, Edge and Brave, and detections can be logged without interrupting people.

    • HighlightAddedGovern

      Sensitive data checks in the browser

      The Sensitive data in prompts rule gains a browser host: with the EndLayer extension, prompts on AI websites in Chrome, Edge and Brave are checked on the device before they are sent.

      Why it mattersThe same data protection policy now covers AI used in the browser, not only coding agents.

      Learn more →
    • AddedGovern

      Log without interrupting

      Log is now an action for sensitive data detectors and custom rules: the action goes through unchanged, no dialog is shown and the detection is recorded in Events.

      Why it mattersYou can measure what would be caught before you start warning or blocking.

    • ImprovedGovern

      Data protection by group

      Sensitive data policy can be assigned to selected device and MDM groups on its own, and device and MDM groups are managed in Governance.

      Why it mattersA pilot group can try a stricter rule before the rest of the fleet.

  10. 5 updates · 1 highlight

    Agent rules, reports and new tools

    The Layer 1 Agent recognises Grok Bot and Grok Build, the Library gains agent rules, and reports come as PDF or slides.

    • HighlightAddedSurface

      Grok Bot and Grok Build recognised

      The Layer 1 Agent detects xAI's Grok Bot desktop app and the Grok Build coding agent. Grok Bot conversations count as usage, and each Grok Bot agent is reported with what it has done.

      Why it mattersComputer-use agents are found and measured like every other AI tool.

      Learn more →
    • AddedGovern

      Rules for what agents may do

      Three rules join the Library: acting on the device, reaching connectors and exchanging files. An agent that does what a rule forbids reads as breaking it. Grok Bot and Grok Build can also be picked in the blocked apps list.

      Why it mattersYou can set what autonomous agents are allowed to do, not only whether they are installed.

    • AddedMonitor

      A page for every autonomous tool

      Each autonomous tool on Monitoring, Agents opens its own page, with its agents grouped by device and their recent runs. Monitoring, Websites shows the AI sites the fleet reaches in a browser.

      Why it mattersYou can see what each agent does and where it runs.

    • AddedGovern

      Reports as PDF or slides

      A report for a chosen period can be produced as a PDF or a slide deck, with an executive summary that lists what needs attention, most urgent first.

      Why it mattersEvidence for leadership comes straight from the console.

    • ImprovedSurface

      Surfaces overview and Google Cloud by organisation

      Surfaces opens on one overview screen, and Google Cloud can be connected for a whole organisation, with each project listed under it. Inbox, Devices gains Approve all for waiting machines.

      Why it mattersYour inventory and approvals are quicker to review at scale.

  11. 3 updates · 2 highlights

    Stop sensitive data reaching coding agents

    The Layer 1 Agent can now stop sensitive data reaching Claude Code and Cursor, and Google Cloud joins the connections.

    • HighlightAddedGovern

      Sensitive data checks in Claude Code and Cursor

      Publish the Sensitive data in prompts policy and the Layer 1 Agent checks each prompt and tool call in Claude Code and Cursor before it goes anywhere. Each detector can record, redact or refuse with a reason, and you can add your own patterns.

      Why it mattersSensitive data is stopped before it reaches the AI, and what leaves the device is the detector and the action, never the matched text.

      Learn more →
    • HighlightAddedSurface

      Google Cloud connection

      Connect a Google Cloud project read-only with a service account. Surfaces now has three kinds: Devices, API and Cloud.

      Why it mattersAI running in your cloud sits in the same inventory as AI on devices.

      Learn more →
    • ImprovedSurface

      Intune rollout to chosen machines

      The Intune rollout can install the Layer 1 Agent on machines you pick, not only on groups, and no longer asks which operating system.

      Why it mattersA pilot can start on a handful of machines.

  12. 4 updates · 1 highlight

    Microsoft Intune, Entra ID and Copilot

    One Microsoft connection now covers Intune, Entra ID, Copilot and Defender, and can roll out the Layer 1 Agent.

    • HighlightAddedSurface

      Microsoft Intune connection

      Sign in with Microsoft to mirror your Intune device inventory, compare it with the devices reporting to EndLayer, and push the Layer 1 Agent to Windows and Mac. One machine can also be enrolled from its device drawer.

      Why it mattersYou can see which managed devices are not yet covered and close the gap from the console.

      Learn more →
    • AddedMonitor

      Entra ID, Copilot and Defender

      The Microsoft connection adds an Entra ID directory mirror with a leaver alert and an audit of AI access, Microsoft 365 Copilot adoption against licensed seats, and Defender threat state on each device.

      Why it mattersIdentity, Copilot use and device threats sit beside the rest of your AI picture.

    • AddedGovern

      Devices can ask to be disconnected

      A person can ask for their machine to be disconnected from its own menu, and an administrator answers the request in the console.

      Why it mattersLeaving the fleet is a recorded decision, not a silent uninstall.

    • AddedMonitor

      Activity map on the home page

      The home page shows the Activity Map at a glance, with a link to the full view.

      Why it mattersHow AI flows across the organisation is visible from the first screen.

  13. 3 updates · 1 highlight

    Newer coding agents, and approved removal

    The Layer 1 Agent recognises Antigravity, DeepSeek Harness and Prime Agent, and blocked apps and MCP servers can be removed once an administrator approves.

    • HighlightAddedGovern

      Removal of blocked apps, after approval

      When a device reports an AI app your organisation blocks, a removal request appears in the console. Nothing is removed until an administrator approves it, and a device can be excepted on its own.

      Why it mattersA blocked tool can be taken off devices, with a person deciding each time.

      Learn more →
    • AddedGovern

      Blocked MCP servers removed

      The Layer 1 Agent removes the MCP servers named in your policy from the tool configs that declare them and reports the outcome. Only the named servers are touched.

      Why it mattersAn MCP server you have ruled out does not stay connected.

      Learn more →
    • AddedSurface

      More agents detected

      The Layer 1 Agent now detects Antigravity, DeepSeek Harness and Prime Agent, and recognises claude.com as an AI domain.

      Why it mattersNewer coding agents show up in the inventory as they appear.

      Learn more →

What comes next is on the roadmap. To see any of this on your own fleet, book a demo.

EndLayer

Give AI room to work.And clear boundaries.

See the Guardian Agent Platform on your own fleet. Find every AI tool, understand what it reaches and costs, and apply the controls your team needs.

One connected layer of control
Devices · Cloud · ApplicationsSurface / Monitor / Govern