Understanding Active Directory: The Technology Behind Enterprise AI Security - Part 5: Browser Configuration and Authentication
Hello! Part 4 covered proxy authentication and access control. This time, we look at what to check in browsers and how to distribute their settings.
The same page may trigger an authentication prompt in one browser or connection path but not another. Before widening the allowed targets or authentication methods, identify who is asking for credentials.
This article primarily covers managed Chrome, Edge and Firefox on Windows, using an explicit forward proxy. The comics are fictional teaching conversations. Configuration examples are illustrative, not records of policy deployment or authentication tests in a live environment.
Series Outline
- Chapter 1: Understanding the basics - Fundamentals of Active Directory and Kerberos/NTLM authentication
- Chapter 2: Building a domain environment - Setting up a lab environment
- Chapter 3: Joining clients and servers to the domain - Domain join in detail
- Chapter 4: Proxy servers and Integrated Windows Authentication
- [★This installment★] Chapter 5: Browser configuration and authentication - Settings for each browser
- Chapter 6: Troubleshooting - Common problems and solutions
- Chapter 7: Security and best practices - Considerations for production environments
- Chapter 8: Practical configuration examples - Integration with AI security tools
5.1 Separate the conditions for integrated authentication
5.1.1 Distinguish site authentication from proxy authentication

- Developer「Why does authentication differ on the same page?」
- Trainer「The site and proxy can ask separately.」
- Trainer「First, check whether it is a 401 or 407.」
- Developer「I will identify the requester before changing lists.」
An unexpected password prompt does not, by itself, identify the setting to change. A website and an intermediary proxy can ask for authentication separately.
| Response | Requester and relevant information |
|---|---|
| 401 Unauthorized | The website. Check WWW-Authenticate and the Authorization sent to that site |
| 407 Proxy Authentication Required | The proxy. Check Proxy-Authenticate and the Proxy-Authorization sent to the proxy |
Negotiate and NTLM can involve several exchanges. An intermediate 407 is not necessarily a failure. Distinguish normal progress from repeated challenges without progress or a final rejection.
Windows zone settings can also affect automatic authentication. On Windows, when the authentication server allowlist is unset, Chrome consults Windows zone decisions about credential use. Treating zones as an Edge-only feature can lead to a wrong diagnosis.
A zone assignment, AD domain membership and possession of Kerberos tickets are separate conditions. Moving a site into a trusted zone will not fix a name or key mismatch. Record the actual proxy route and authentication requester before reviewing the allowed scope.
5.1.2 Align the connection name with DNS and the SPN

- Developer「Does the same IP mean identical authentication?」
- Trainer「The connection name also affects the SPN.」
- Trainer「Align DNS, service names and allowed targets.」
- Developer「Switching to an IP is not proof of a fix.」
Reaching the same server does not guarantee the same authentication conditions. A short name, FQDN, alias and IP address can produce different name-resolution and authentication behavior.
The authentication target depends on who sent the challenge. For a website’s 401, check the site connection name and SPN. For a 407 from an explicit forward proxy, check the proxy connection name and SPN. Do not confuse the destination URL with the proxy’s name.
| Connection name | What to check |
|---|---|
| A short name such as intranet | The DNS suffix used for resolution and the automatic authentication decision |
| intranet.ad.example.com | Alignment of the connection name, certificate, allowed target and AD service name |
| Alias or CNAME | The SPN actually constructed, including DNS canonicalization and port-related policies |
| IP address | Windows does not attempt Kerberos for an IP name by default. Success using another scheme does not prove equivalent authentication |
The SPN matters for Kerberos HTTP services. Policies such as DisableAuthNegotiateCnameLookup affect CNAME handling, while EnableAuthNegotiatePort affects port inclusion. Check the SPN actually constructed and its registered account rather than relying on the resolved IP alone.
Windows provides an exception for IP-based SPNs. Microsoft presents it as an option when the application cannot be changed to use a DNS name. Here, the baseline is the administrator-approved DNS name, rather than broadening exceptions.
The presence of the word intranet somewhere in a URL does not prove the target is trusted. Specify managed names and avoid wildcards that inadvertently include external hosts.
5.1.3 Distinguish browser settings from credential delegation
Permitting authentication to a server and delegating a user’s credentials are separate decisions. Do not add delegation targets indiscriminately to fix an ordinary sign-in.
| Browser and setting | Purpose |
|---|---|
| Chrome / Edge AuthServerAllowlist | Allows integrated authentication to specified web servers. Separate multiple names with commas |
| Chrome / Edge AuthNegotiateDelegateAllowlist | Allows credential delegation. When unset, delegation is not permitted. Design it separately when needed |
| Firefox Authentication.SPNEGO | Defines sites that may use Kerberos |
| Firefox Authentication.AllowProxies | Controls proxy authentication separately from the site SPNEGO list |
Chrome and Edge distinguish a proxy challenge from a challenge by an allowed web server. Adding a proxy name to AuthServerAllowlist is not a universal fix. Check the route, advertised schemes and available credentials too.
The following minimal Firefox policies.json example configures site authentication using an example name. The HTTPS target and its descendant hosts are covered, so confirm the management scope of the subdomains as well. This is not a proxy configuration example.
{
"policies": {
"Authentication": {
"SPNEGO": [
"https://intranet.ad.example.com"
]
}
}
}The example does not add Delegated, because it does not require delegation. AllowProxies separately controls proxy behavior. Also distinguish HTTP authentication from SAML or OpenID Connect web sign-in and from IE-mode compatibility settings.
5.2 Distribute settings safely to managed browsers
5.2.1 Chrome does not universally disable integrated authentication by default

- Developer「Is automatic authentication always off in Chrome?」
- Trainer「No. That blanket statement is incorrect.」
- Trainer「Check the OS, target, policies and browsing mode.」
- Developer「I will compare the actual conditions.」
It is inaccurate to say that Chrome disables Windows Integrated Authentication by default in all cases. Separate the unset-policy behavior, operating system, target and browsing mode.
The current Chrome administrative template states that, when the authentication server allowlist is unset, Chrome determines whether the server is on the intranet. On Windows, Negotiate uses SSPI. Chrome is not isolated from Windows authentication facilities.
Since Chrome 81, ambient authentication using default credentials is disabled by default in Incognito and Guest modes. AmbientAuthenticationInPrivateModesEnabled controls that behavior. Comparing a private window with a regular window is not a reason to enable ambient authentication in every mode.
- 1Compare the same device, user, URL and proxy route.
- 2Match browser and OS versions and distinguish regular from private browsing.
- 3Inspect policy values, sources and errors.
- 4Check required tickets, service names and advertised authentication schemes.
- 5Once the difference is understood, apply only the necessary changes to a pilot group.
Similarly, do not describe Firefox as incapable of authentication by default in every environment. Check site allowlists, proxy settings, private mode and OS authentication facilities. A controlled comparison is more useful than guessing internal behavior from a browser’s name.
5.2.2 Verify delivery from the GPO to the browser

- Developer「Is creating the GPO enough?」
- Trainer「Check that the device and browser receive it.」
- Trainer「Check values, sources and errors, then restart.」
- Developer「I will test both access and rejection.」
Creating a GPO and delivering its settings to a user’s browser are separate steps. Start with a small, reversible pilot for selected devices and users.
- 1Obtain official Chrome, Edge and Firefox administrative templates and check compatibility with the deployed versions.
- 2For domain administration, inspect the existing Central Store. Version the ADMX files and matching language ADML files; do not overwrite shared templates without planning.
- 3Select the OU, user or computer scope, GPO links and filters.
- 4Configure only the names that need authentication. Treat delegation separately, after confirming its purpose and targets.
- 5Read the client’s resulting policy, restart the browser where required, and verify behavior.
The Central Store holds shared templates under SYSVOL. Editing only the management workstation’s local PolicyDefinitions does not necessarily change an editor that uses the Central Store. Placing an ADMX file does not itself change client settings.
The following example is for a Windows test device. gpupdate /force reapplies all policies, so consider its effect on other settings. Change the report filename or path as appropriate.
gpupdate /force
gpresult /h ad-browser-policy-report.htmlgpresult reports Group Policy results. In the browser, inspect values and errors at chrome://policy, edge://policy, or Firefox’s about:policies. In Chrome, also inspect the policy source to distinguish GPO from cloud management.
Reloading policies is not always sufficient. Some settings require a restart; Edge’s authentication server and delegation allowlists are examples. Test rejection of disallowed targets and unauthenticated access as well as successful access.
5.2.3 Separate password storage from authentication caches

- Developer「Does a shorter cache erase saved passwords?」
- Trainer「Server caches and client storage are separate.」
- Trainer「The client may send credentials again.」
- Developer「I will check each place that retains state.」
Shortening a proxy’s cache lifetime does not erase passwords stored in a browser. Identify what each component retains and for how long.
| Retained information | What to distinguish |
|---|---|
| Browser-saved passwords | Preventing new saves and handling existing entries are separate matters |
| Browser HTTP authentication state or connections | Reuse cannot always be explained by the password manager alone |
| OS credentials or Kerberos tickets | Changing the browser profile’s display name does not replace them |
| Squid Basic authentication cache | credentialsttl governs server-side revalidation. It does not clear client memory or saved passwords |
With Edge’s PasswordManagerEnabled disabled, previously saved passwords remain usable. Do not report that existing credentials were removed merely because new saves were disabled. Verify the HTTP authentication prompt and its saving behavior for the particular browser version.
On shared devices, verify the authenticated account after a handover, including OS sign-out and sign-in, browser processes, connections and saved state. This article does not recommend creating a custom extension that injects credentials into every URL to prevent password saving.
If Basic is used, protect the client-to-proxy connection, not just the connection to the destination website. An ACL that checks whether the destination uses HTTPS does not guarantee encrypted transmission of the proxy credentials.
5.2.4 Define exceptions and roll out in stages

- Developer「Does a VPN guarantee integrated authentication?」
- Trainer「A VPN alone does not meet every requirement.」
- Trainer「Choose the devices and path, then pilot them.」
- Developer「I will roll out tested, limited exceptions.」
Office, VPN and personal-device categories do not determine the authentication method on their own. Connectivity is not enough if the required tickets, names or client support are missing.
| Test scenario | Expected result to verify |
|---|---|
| Regular managed device | The intended account reaches the necessary destinations |
| VPN or another network | DNS, proxy routing and necessary authentication traffic work together |
| User handover on a shared device | The previous user’s credentials are not reused |
| Unmanaged device or unsupported client | The documented guidance or rejection occurs, rather than an unplanned weaker fallback |
| Disallowed site or expired exception | Credentials are not sent unnecessarily and old exceptions do not remain |
Do not automatically select Basic merely because a client uses a VPN or is not domain-joined. If an exception is needed, define its users, devices, protected connection, expiry and revocation process, then test with the supported client and proxy.
Tell users which browser and official address to use and where to report suspicious prompts. Diagnostic logs may include URLs, account names and tokens. Remove sensitive information before sharing them.
When considering passwordless authentication or another SSO platform, remember that a web sign-in feature does not automatically replace a proxy’s 407 authentication. Product support and the authentication path need their own design. Part 6 covers troubleshooting when authentication still fails after the conditions have been aligned.
Company-wide AI security, from foundations to operations.
From foundations such as identity and access management to LLM-specific risks (prompt injection, data leakage, harmful output)—using AI safely across an organization involves a wide range of issues.
We develop and provide our own LLM audit product, "LLM-Audit." Drawing on that expertise, we provide hands-on support from audits and assessments to defenses and secure operational design.
Explore our LLM security services →