Skip to main content
Each entry below is one release, newest first. Agent releases are named Endpoint with their version. Console and platform releases are named Platform with their date.
Console and platform, 19 November 2026
Two releases went live on 19 November. They add a watch only mode for a first look at AI use, personal data found on the laptop without acting on it, better Swiss detection, and a clearer notice for your people. They work with every agent from 1.6.0. Features that need a newer desktop app say so.
  • Discovery only. One reversible switch makes a workspace watch and nothing else: it records which AI services people use, nothing is blocked, no prompt is changed and no prompt text is sent. Choose it in the Default policy or as Watch only (passive) in Setup, and turn it off to put back what it changed. See Discovery only.
  • Personal data found on the laptop. With Discovery only on, devices on desktop 2.1.12 or later look for the personal data types you chose and report the category and a count. Nothing is changed and no text is sent. See Personal data found on the laptop.
  • Impersonating sites and lookalikes. The notice that blocking is off now comes first on every page for the administrators who can act on it. A site listed by mistake can be marked Not a lookalike, and a lookalike is listed as soon as a device reaches it. See Impersonating AI sites and lookalikes.
  • Swiss detection. Swiss phone numbers, likely AHV numbers (read as Likely AHV) and passwords written in German, French and Italian are found and masked. See Swiss numbers and passwords.
  • Files are redacted, not blocked. A file that still holds personal data after the device cleaned it is no longer blocked for that reason by default, and with desktop 2.1.13 and later the device redacts what was found. A file’s name is not kept when the workspace keeps no prompt text or Discovery only is on. See Files with personal data.
  • Removed devices read as removed. A device the person uninstalled themselves, on desktop 2.1.12 or later, reads Removed by the user in the device list instead of staying protected. See What the console shows after a removal.
  • Employee notice in five languages. The notice your people read on the laptop is available in English, German, French, Spanish and Portuguese, says where text goes, and sends rights requests to your administrator. See Employee notice.
  • Setup. Setup remembers the policy drafts it started, says when the App it used was deleted, and offers Watch only (passive). See What Setup remembers.
  • Countries in one place. Settings → Countries is where you change the countries your workspace operates in. See Countries.
  • Traffic drift is information. A shift in traffic is shown as something to review and never as a critical alert.
  • Text areas. Long text boxes open at a height that fits their purpose and grow as you type, and fit a laptop screen.
Console and platform, 18 November 2026
This release adds the Runtime Security dashboard, a first setup for Compliance, more ways to add your company’s own AI services, a proof of the machine key when a device enrols again, and a redesigned Settings. It works with every agent from 1.6.0. Features that need a newer desktop app say so.
  • Runtime Security dashboard. The page opens on a sentence that says whether your Apps are protected and receiving, then the numbers for the window you pick, what was caught, what changed and next steps. It counts Apps the way the Apps list does. See The dashboard.
  • Apps list. A row opens its App, and the Apps that need you come first. See The Apps list.
  • Compliance first setup. With no framework switched on, Compliance asks where you operate, then only the questions those countries raise, then shows what switches on and why. Nothing is saved until the last step. See First setup.
  • Your company’s own AI services. Devices accept your company’s certificates only when an administrator allows it, an administrator can let Windows devices read internal services on private addresses, and you can add your own Azure OpenAI, Azure AI Foundry or Amazon Bedrock endpoint by its exact name. Each choice is recorded in the audit trail. See Company certificates, Internal services on private addresses and Cloud model endpoints by exact name.
  • Importing a policy is a main way to add one. Endpoint → Policies opens with Import the policy you already wrote beside Write one here, and an import never proposes a control the document did not state. See Import a policy.
  • Settings redesigned. Every Settings page uses the same rows, switches and choices. Integrations is shown only where Data Security is deployed and licensed. Global reviewers are added and removed one at a time, so two admins editing at once both keep their change, and the API gains POST /api/settings/reviewers/{user_id} and DELETE /api/settings/reviewers/{user_id}. See Global reviewers.
  • Machine key proof when a device enrols again. A device with a machine key proves it holds the key before its enrolment is accepted. Workspaces watch and record by default, and enforcement is switched on per workspace. See Machine key proof.
  • Refuse when the app cannot be named. A Default policy switch that makes a block rule with no host also refuse traffic from an application the device could not name. It is off by default, and the policy shows how many connections devices could not name. See Refuse when the app cannot be named.
  • Topic rules on every App. An App’s own topic rules work whichever personal data types the App has switched on.
  • Warn instead of block for lookalike sites. In the Default policy, When a person opens one lets you choose Warn: the person sees a page saying the site is not the service it imitates, with a link to the real one, and may still continue. It needs desktop 2.1.8 or later. See Block or warn.
Console and platform, 17 November 2026
This release rebuilds Runtime Security Apps around three protections, adds a client guide for chatbots, makes every view in the console linkable, and adds replace and delete for SCIM groups. It works with every agent from 1.6.0. Features that need a newer agent say so.
  • Apps rebuilt around three protections. Each App shows personal data, topics and prompt injection on one line each, with what it did in the window and Change to edit it in place. Personal data can be pseudonymized and restored, topics are the App’s own subjects to block or flag, and prompt injection is blocked or flagged only. See The three protections.
  • New Apps list and pages. The list shows the three protections side by side. An App’s page has Overview and Configuration, New app sets the protections as it creates the App, and Connect gives a scan key limited to the App and a test request. See Apps.
  • Chatbot integration. For an App set to pseudonymize and restore, a value keeps its placeholder in every message of a conversation, restoring never puts back a value the model added on its own, and /scan/restore puts values back into text. An App set to flag only answers flag when it catches an attack, and redact with also_flagged when the same text also holds personal data. See Protect a chatbot.
  • Clearer errors and counts. Errors carry a code and, where it applies, a wait in Retry-After. They never repeat what was sent. A request that ends in 503 is not counted against your allowance, for single prompt, reply and file scans, and an App appears as receiving as soon as it has received traffic. The SDKs read the code and the wait, and retry a 429 or 503 a bounded number of times. See Errors.
  • Keys go with their App. Archiving or deleting an App revokes every key limited to that App alone, and each is recorded in the audit trail. See Keys that belong to an App.
  • Find your way. Every view has its own address, the breadcrumb names the record and the tab, and a link that takes you into another product offers the way back. See Finding your way.
  • SCIM groups. Your identity provider can replace and delete a group. A group that grants platform roles is refused on delete, and Endpoint groups keep their policy assignments. See Groups.
  • Policy import. An import never turns a sentence into a control the document did not state. See An import never proposes a control the document did not state.
  • Lookalike sites. A site dressed as a known AI service is listed as an impersonation, and is not counted as an AI service on the Overview, the AI services page or in reports. See Sites that impersonate an AI service.
  • Daily request allowance. If your plan has one, requests well beyond it are paced and never refused for the day: you get 429 with DAILY_ALLOWANCE_SOFT_LIMIT and a wait. See Scan API.
Console and platform, 16 November 2026
This release redesigns Compliance and Overview, brings one list design to the console, and adds lookalike AI site detection, an auditor pack with read-only links, and collect diagnostics for a device. It works with every agent from 1.6.0. Features that need a newer agent say so.
  • Compliance redesigned. Each framework shows a coverage bar, a week by week history you can rewind, next steps, and a panel for every control. See Coverage, history and next steps.
  • Evidence records. Record a document for an organisational control, have a second person sign it off, and withdraw it before it is signed off. See Evidence records.
  • Auditor pack and read-only links. Download one framework and period as a ZIP, and make an expiring read-only link an auditor opens without an account. A link stops for good when Compliance or that framework is turned off. See Auditor pack and shared links.
  • Overview redesigned. What changed against the period before, one rule for counting devices, and setup progress that matches the sidebar. See The Overview page.
  • Setup in place. Setup steps open where you are instead of sending you elsewhere. See Setup opens in place.
  • Features per policy. A policy says whether it applies to Endpoint, Microsoft 365 Copilot, AI services, or SDK and API scans, and the Default policy has a switch for each. See Features a policy applies to.
  • One list design. Lists across the console share one toolbar, open a row in place, and keep the actions in the open row.
  • Lookalike AI site detection. Sites that copy a known AI service are listed with what they imitate, and the clearest can be blocked. Blocking is off for existing workspaces and on for workspaces created from now on. An administrator turns it on from the notice Turn on blocking or in the Default policy. See Blocking lookalike sites.
  • A shared AI catalogue. More AI services are recognised over time without an update on your side. See The catalogue grows.
  • Reach for AI apps added by hand. A service you added says whether devices reached it and could read it. See Reach of a service you added.
  • Collect diagnostics. An administrator can collect diagnostics from a device with no step for the person. It requires a recent desktop version. See Collect diagnostics.
  • Clearer device problems. A device that is not protected describes the cause in plain words. See Why a device reads Partly protected.
  • Privacy. A workspace that does not show names never shows them in device causes, code signers or package scopes.
  • Safer file previews. File previews are served with stricter security headers.
  • Settings for newer desktop apps. The console marks a setting that only newer desktop apps read with the version it needs: updates (2.1.6), the block page message and contact, the redaction notice and reporting names looked up by any program (2.1.7). See Settings that need a newer desktop app.
9 November 2026
Endpoint 2.1.5 for Windows and Mac collects the changes since 2.0.0. It is on the stable channel, so devices on 2.0.0 update automatically. Features that need a newer platform say so.
  • Stronger device identity. A device proves it is the same device when it enrols again, on Windows and Mac. A clean uninstall removes the device identity, so a reinstall needs the enrollment token. See Uninstalling the agent.
  • Shadow AI by DNS lookups. The agent notices AI services by the names a device looks up. It reads names only. The workspace setting Notice AI services by DNS lookups controls it. See Notice AI services by DNS lookups.
  • Microsoft 365 Copilot. Only what a person types is inspected, once. History, titles and type-ahead suggestions are no longer recorded as prompts, and a refused turn ends with a plain message.
  • Protections follow the console. The agent applies every Protections setting per workspace, policy and AI service. A data type set to Block is now refused on the device, where before it could be sent masked.
  • More personal data on the device. Mexican RFC, CURP and CLABE numbers are recognised on the device, with their check digits.
  • Block events. Blocks by host or category are reported as events with no prompt text. They appear in the console with Platform 2026-11-09.
  • Steadier operation. A short outage or a rate limit is retried instead of failing open, the last good licence and policy are kept, and the app starts reliably after an upgrade or a restart.
  • Regex host rules. A regex host rule is matched without regard to case on every route. A rule that used \S, \D, \W or \B and never matched may start matching, so review those rules.
  • Diagnostics. Automatic diagnostics upload is off unless the workspace turns it on. Diagnostics an administrator requests still work.
Console and platform, 9 November 2026
This release adds a preview before every Intune change, simpler Intune connection and permissions, policy fixes for the Wintun block, a setting for noticing AI services by DNS lookups, and a structured data form of syslog export. It works with every agent from 1.6.1. Features that need a newer agent say so.
  • Every Intune change is previewed. Before Blindsight writes to your Intune, the console shows what it will create, change, assign or remove, what it leaves untouched and how to undo it, and writes only what you confirmed. See Every change is previewed first.
  • One Microsoft approval to connect. Connecting Intune asks Microsoft once instead of twice. See Connecting with one Microsoft approval.
  • Approve more access. When an optional Microsoft permission is missing, the Intune page offers Approve more access. Existing tenants approve the new read permission DeviceManagementConfiguration.Read.All once to enable policy checks. See Approve more access.
  • Policies redesigned. Each policy is one page in sections: Overview, Personal data, AI services, Topics, Files, Assignments and History. The Default policy has the same shape. Protections, Default policy, Assignments and Import a policy now live under Policies, and the old addresses redirect. See One page per policy.
  • Assign and check. Assign a policy to devices from its Assignments section, see every assignment in All assignments, and use Check a person to see what applies to a person and why, including data types that differ from the default and the policy each comes from. History is read-only. See Check a person.
  • Blocks appear as events. Blocks stopped on the device now show in the console as events, with no prompt text.
  • Intune fix card. An applied fix stays listed with Undo after devices recover, and a failed write offers Check again. See Policy fixes for the Wintun block.
  • Policy fixes for Wintun. Check one group, preview the policy, apply it and undo it, from the Wintun notice. It needs the Blindsight Intune fixes approval. See Policy fixes for the Wintun block.
  • Notice AI services by DNS lookups. A new Default policy setting, on by default, for agents 2.1.4 and later. See Notice AI services by DNS lookups.
  • Syslog structured data. RFC 5424 with structured data is a separate format, blindsight@67580, beside the unchanged key=value form. See Syslog export to your SIEM.
  • Places default to Keep. Place names such as countries and cities are no longer pseudonymized unless an administrator sets Places to Protect or Block. Postal addresses, email addresses and card numbers are unaffected.
  • Events lists. The lists offer a relevant only view, with options to include or show only hidden records and a count of each.
  • Install page. Each enrollment token shows its limit and expiry.
  • Endpoint without Runtime Security. A workspace licensed for Endpoint alone uses Endpoint fully.
  • A quieter top bar. Data Security appears only when your workspace is licensed for it. The request and latency figures are gone from the top bar, with their Appearance setting, and the background activity indicator shows only while work is running.
Console and platform, 6 November 2026
This release adds native syslog export of Endpoint events to your SIEM, topic rules in Endpoint policies, a more complete policy import, and links between devices and the people who use them.
  • Syslog export. Send blocked, redacted and flagged prompts, policy changes and device protection changes to your SIEM’s syslog collector, in CEF, JSON or RFC 5424, over TLS by default. Send a test event before you start, and follow delivery health on the destination’s row. The text of a prompt is sent only when you turn on Include what was typed. See Syslog export to your SIEM.
  • Topic rules. Describe a subject in a sentence, and choose to flag or block prompts about it, whatever words people use. Topic rules are part of a policy and reach everyone the policy is assigned to. See Topic rules.
  • Policy import. An import now maps named AI services, categories of AI services, file scanning and per-service settings to policy, and lists every sentence it cannot turn into a rule. See Import a policy.
  • Device linking and person records. Laptops are linked to the people who use them from Intune, so a policy assigned to a person reaches their device. People without a console account get a person record that you can assign policies to. Related notices are grouped, and a closed notice moves to Dismissed notices. See Device linking and person records.
2 November 2026
Endpoint 2.0.0 for Windows and Mac makes inspection more responsive under heavy AI use, broadens what Shadow AI discovers, and simplifies enrolment on Mac. Devices on 1.6.x update automatically from the stable channel.
  • Responsive under load. Prompt checks run alongside the connection instead of holding it up, so inspection stays smooth when people use AI heavily.
  • AI traffic only on Windows. The default mode routes only AI traffic through Blindsight. Downloads and everyday traffic are unaffected.
  • Broader inspection coverage. More AI tools, including coding assistants, are inspected by default.
  • Broader Shadow AI discovery. More of the AI services your people reach are recognised and shown in the console.
  • Mac enrolment from device management. A Mac can enrol with a deployment token delivered by your device management.
  • Clearer messages. Fewer false alarms, and setup errors are explained in plain language.
Console and platform, 2 November 2026
This release brings a new way to govern AI services, finer inspection settings, clearer device protection states and a defined licence lifecycle. It works with every agent from 1.6.1. Features that need a newer agent say so on their page.
  • AI services, Catalogue and Observed. Endpoint → AI services lists every recognised AI service and what your fleet uses. Shadow AI now lists unapproved services only, and approving a service keeps it inspected. The most-used AI services are inspected by default, so figures on Overview may change on the day. See AI services.
  • Inspection settings per service and per policy. Pseudonymization and prompt-injection detection can be set to Inherit, On or Off for a service or within a policy.
  • Custom AI services. Add an internal or unlisted AI service by its domain, choose what is read and who may use it, and confirm it with a test prompt. See Custom AI services.
  • Clearer protection states. Devices read Protected, Partly protected, Not protected or Starting, with a short reason. The Intune page groups devices the same way, a single notice names the setting to change for devices that need the Wintun adapter allowed, and a Partly protected alert keeps you informed. See Matching and the figures.
  • Licence lifecycle. When a licence ends, the workspace moves through notice, read-only and export phases, with an email to administrators at each step. See When a licence ends.
  • Simpler navigation. Settings, Access management and Profile show their sections in the main sidebar. A figure that cannot be read reads Not reported, never zero. See Settings.
29 October 2026
Endpoint 1.6.6 improves how the Windows agent handles device installation restrictions, shared machines and company proxies.
  • Steadier protection under device installation restrictions. Devices where a policy restricts the Wintun adapter report their state consistently from the moment they start.
  • Machines with nobody signed in report their true state and no longer raise false alerts.
  • Faster recovery. When an administrator allows the Wintun adapter, the agent picks it up sooner.
  • Company proxies set for apps. A company proxy configured only in the Windows connection settings used by PowerShell, .NET and Office is now detected and routed through Blindsight.
27 October 2026
Endpoint 1.6.5 strengthens the Windows agent on networks where traffic goes through a company proxy and inspection through the company proxy is switched on.
  • The proxy route is maintained. The agent keeps traffic routed through Blindsight to the company proxy, including an automatically discovered proxy, and restores the route when it is changed, for example by a Group Policy refresh.
  • Accurate state. A device without the route in place reads Partly protected with the reason.
  • Pause and revoke. A paused or revoked device gets its own proxy settings back. Only changes the agent did not make are reported as tamper.
  • Quicker start on networks where inspection through the company proxy is switched off.
26 October 2026
Endpoint 1.6.4 for Windows, with the platform release it ships with, focuses on reliable installs and upgrades, full support for company proxies, stronger tamper protection and a new Intune connection.
Replace your alert destination URLs. Slack, Teams, webhook and Splunk destination URLs created before this release should be replaced. Create new ones at the provider and edit each destination to use them. See Alert destinations.
Before you update
  • Update your Intune detection rules. Detect the agent by the file version of BlindsightAgent.exe, greater than or equal to 1.6.4, or select Update in Connect Intune to write the new rules. See Deploying with Intune.
  • Inspection through a company proxy is on by default for workspaces that never chose a value. See Proxies.
  • Add the 1.6.4 allow rules if you use App Control or AppLocker. See Locked-down PowerShell and App Control.
Highlights
  • Reliable installs and upgrades. Installs report a clear result, upgrades on Windows Server are dependable, and an interrupted install repairs itself.
  • Company proxies. Enrolment, sign-in, updates and scans follow static, PAC and WPAD proxies, including NTLM and Kerberos authentication and TLS inspection products.
  • Stronger tamper protection. Stopping, disabling or renaming the agent raises a tamper alert, and a removed inspection certificate is restored and reported.
  • Connect Intune. Connect a tenant, see which devices run the agent, and deploy, update or remove it from the console. See Connect Intune.
  • Security. Alert destination URLs are masked throughout the console and audit trail, and proxy credentials are never written to disk.