Claude Code admin settings are changing so Claude Code Desktop reads the same managed admin policy as the CLI and IDE from server-managed settings, MDM policy, or managed-settings.json, using the same precedence rules. The most urgent change is that the Claude Code Desktop admin toggles for Allow bypass permissions mode and Allow auto permissions mode are being retired. Unless your organization has already explicitly disabled those modes, they are available by default; admins who want them disabled need to add the matching managed policy by June 5, 2026.
This matters for IT, security, engineering, RevOps, and business leaders because Claude Code can read code, edit files, run commands, call tools, connect to MCP servers, and operate across local development environments. Vantage Point helps organizations turn AI adoption into practical governance: clear settings, defined workflows, secure data access, and change management that supports productivity without creating unmanaged risk.
managed-settings.json./status and /permissions.Anthropic notified Claude Code administrators that the Claude Code admin settings page has been reorganized and that Claude Code Desktop will now read the same admin policy as the CLI and IDE. That policy can come from managed settings, MDM controls such as macOS plist or Windows registry, or managed-settings.json, with the same precedence model.
The practical change is simple: Claude Code administration is becoming more unified across surfaces. Instead of treating Desktop settings as a separate admin control point, organizations should think in terms of one managed policy model for Claude Code.
The urgent item is the retirement of two Desktop admin settings toggles:
Anthropic’s notice says these modes are available by default unless your organization has already explicitly disabled them. To disable them going forward, admins need to add the matching policy to their organization’s Claude Code settings by June 5, 2026.
Source: Anthropic Claude Code settings documentation and the Anthropic admin email notification provided to Vantage Point.
Claude Code admin settings matter because AI coding tools are no longer small productivity add-ons. They are becoming operational systems that can touch repositories, local files, development commands, package managers, terminals, APIs, MCP servers, plugins, hooks, and shared team workflows.
That is powerful. It also means governance cannot be limited to “who has a license?” Leaders need to answer deeper questions:
This is the same operating-model challenge Vantage Point often sees with Salesforce, HubSpot, AI, and integration programs. The tool is not the strategy. The strategy is the combination of workflow design, data governance, security controls, adoption planning, and ongoing support.
For organizations building AI into customer operations, Vantage Point connects this work to AI-driven personalization and analytics, compliance and security solutions, and advisory and change management.
Claude Code managed settings are organization-controlled configuration policies that take precedence over user, project, and local settings. They let administrators define what Claude Code can and cannot do across users and machines.
Anthropic documents several configuration scopes:
| Scope | Typical location | Who controls it | Business use |
|---|---|---|---|
| Managed | Claude admin console, MDM, registry, or system-level managed-settings.json |
IT, security, platform administrators | Organization-wide policy enforcement |
| User | ~/.claude/settings.json |
Individual user | Personal preferences and global defaults |
| Project | .claude/settings.json |
Team or repository maintainers | Shared project settings committed to source control |
| Local | .claude/settings.local.json |
Individual user | Machine-specific or experimental project settings |
The most important principle is precedence. Managed settings sit at the top. They cannot be overridden by user preferences, project settings, or command-line arguments.
That is why the June 5 change matters. If an organization wants to prevent auto mode or bypass permissions mode, the safest approach is to enforce that choice through managed settings rather than relying on user behavior or one-off Desktop toggles.
Auto mode and bypass permissions mode are Claude Code permission modes that reduce or remove approval prompts. They are useful in the right context, but they should not be enabled casually across sensitive environments.
| Mode | What it does | Best use | Governance concern |
|---|---|---|---|
| Default | Prompts before tool actions that need approval | Sensitive work, early adoption, general use | Lower productivity for repetitive tasks |
| Accept edits | Allows file edits and common filesystem commands in working directories | Iterating on code with human review | Still requires clear protected path and repo hygiene |
| Plan | Lets Claude research and propose before editing | Architecture review, codebase exploration, risky changes | Teams must actually review the plan before execution |
| Auto | Runs with background safety checks instead of frequent prompts | Longer trusted tasks where direction is clear | Still a research-preview-style autonomy model and not a substitute for governance |
| DontAsk | Only pre-approved tools run; other prompts are denied | CI, scripts, locked-down automation | Requires well-designed allow rules |
| Bypass permissions | Skips permission prompts and safety checks | Isolated containers or VMs only | High risk on normal workstations or production-connected environments |
The big distinction: auto mode still applies safety checks, while bypass permissions mode skips permission prompts and safety checks. Bypass permissions mode should generally be limited to isolated environments, such as containers or virtual machines where Claude Code cannot damage host systems or access sensitive resources.
Admins should decide whether auto mode and bypass permissions mode are appropriate for the organization, then encode that decision in managed settings before the June 5 deadline.
A practical action plan looks like this:
managed-settings.json depending on your device environment./status and /permissions in Claude Code to confirm the active settings source and permission rules.The specific JSON should always be validated against Anthropic’s current documentation and tested in a pilot group. As a starting point, teams should review these controls.
| Control area | Why it matters | Example setting or concept |
|---|---|---|
| Disable bypass permissions | Prevents users from skipping permission prompts and safety checks | permissions.disableBypassPermissionsMode |
| Disable auto mode | Prevents promptless auto mode when the organization is not ready for it | permissions.disableAutoMode |
| Managed-only permissions | Prevents users or projects from defining their own allow, ask, or deny rules | allowManagedPermissionRulesOnly |
| Deny sensitive reads | Blocks access to .env, secrets, credentials, and sensitive files |
permissions.deny rules |
| Restrict Bash/network paths | Reduces risk from shell commands, curl, wget, and external calls |
Permission rules plus sandbox settings |
| Control MCP servers | Prevents unmanaged tools and integrations from entering the environment | allowedMcpServers, deniedMcpServers, allowManagedMcpServersOnly |
| Restrict plugin marketplaces | Limits extensions to approved sources | strictKnownMarketplaces, blockedMarketplaces |
| Sandboxing | Adds OS-level filesystem and network boundaries around Bash commands | sandbox.enabled, network/domain controls |
| Version floor | Ensures users run a minimum Claude Code version with required controls | minimumVersion |
A conservative managed settings pattern may look like this conceptually:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(./secrets/**)",
"Bash(curl *)",
"Bash(wget *)"
],
"disableBypassPermissionsMode": "disable",
"disableAutoMode": "disable"
},
"allowManagedPermissionRulesOnly": true
}Important note: Organizations should test this in a non-production pilot before broad deployment. The exact policy should reflect how your teams build, test, deploy, and use approved internal tools.
Anthropic documents several ways to deliver managed settings. Each option fits a different operating model.
| Delivery method | Best for | Notes |
|---|---|---|
| Server-managed settings | Teams and Enterprise customers that want centralized web-based policy | Delivered through Claude.ai admin settings; requires supported Claude plan and network access |
| MDM / OS-level policy | Managed macOS and Windows fleets | Stronger endpoint enforcement because policy is deployed by IT/device management |
| File-based managed settings | Linux, WSL, servers, developer images, or managed workstations | Uses system directories such as /etc/claude-code/ or platform equivalents |
| Windows registry policy | Windows enterprise deployments | HKLM is stronger than HKCU because it requires admin privileges |
The main architecture decision is whether your organization manages devices centrally. If yes, endpoint-managed settings through MDM or registry controls may provide stronger assurance. If not, server-managed settings are often faster to deploy and easier to administer.
For mixed environments, teams may need both: server-managed settings for Claude.ai users and endpoint-managed policies for machines, providers, or environments that cannot rely on server-managed delivery alone.
Teams should govern Claude Code permissions using defense in depth: permission rules, sandboxing, managed settings, user training, and monitoring.
Permission rules control what Claude Code can attempt. Sandboxing controls what subprocesses can actually reach at the operating-system level. That distinction matters. A deny rule for WebFetch can block Claude’s web fetch tool, but if Bash is broadly allowed, a command like curl may still reach external endpoints unless Bash/network behavior is also restricted.
A practical governance model includes:
.env, credential files, secrets folders, private keys, exports, and customer data extracts.Bash(npm test) instead of broad rules such as Bash(npm *).This is where AI governance overlaps with broader operating design. Vantage Point often helps teams connect AI controls to managed services and ongoing support, workflow automation and process optimization, and system integration and data migration planning.
If teams ignore the update, they may unintentionally leave higher-autonomy Claude Code modes available across Desktop environments. That does not automatically mean something bad will happen, but it does mean the organization’s actual AI operating model may be looser than its intended policy.
Common risks include:
| Risk | What it looks like | How to reduce it |
|---|---|---|
| Unapproved autonomy | Users enable auto or bypass behavior without a defined policy | Disable modes through managed settings or explicitly document approved use cases |
| Secret exposure | Claude Code reads or processes .env, tokens, credentials, or sensitive exports |
Add deny rules and sandbox boundaries for sensitive paths |
| Unsafe shell commands | Broad command rules allow scripts, network calls, or destructive operations | Use narrow allow rules, deny risky commands, and test policy behavior |
| Unmanaged integrations | Users add MCP servers or plugins from unreviewed sources | Use MCP allowlists and marketplace restrictions |
| Inconsistent rollout | CLI, IDE, and Desktop behave differently because policies are not unified | Standardize managed settings across surfaces |
| Adoption backlash | Controls are imposed without explaining the workflow impact | Pair policy changes with enablement, examples, and exception paths |
The goal is not to block Claude Code. The goal is to make Claude Code safe enough to scale.
CRM, RevOps, and data teams should think about Claude Code as part of the broader AI operations stack, not just as an engineering utility. Many business teams increasingly rely on technical workflows: API scripts, migration utilities, data cleanup jobs, reporting automation, integration mappings, and configuration-as-code patterns.
That creates practical questions:
These questions sit squarely at the intersection of CRM strategy, AI governance, and implementation operations. For teams using Salesforce, HubSpot, middleware, or custom data pipelines, the safest approach is to align Claude Code policies with the same data governance standards used across CRM and integration programs.
Vantage Point helps organizations evaluate, implement, and optimize Salesforce and HubSpot based on their operating model, data needs, adoption goals, and growth strategy. That same lens applies to AI coding tools: define the work, define the data, define the controls, then scale adoption.
Use this checklist before June 5 and as part of your ongoing AI operations review.
| Checklist item | Owner | Status |
|---|---|---|
| Confirm whether Claude Code Desktop, CLI, and IDE are in use | IT / Engineering | Not started / In progress / Done |
| Decide whether auto mode is allowed | Security / Engineering leadership | Not started / In progress / Done |
| Decide whether bypass permissions mode is allowed | Security / Engineering leadership | Not started / In progress / Done |
| Deploy managed settings before June 5 | IT / Platform team | Not started / In progress / Done |
Add deny rules for .env, secrets, credentials, and sensitive exports |
Security / Platform team | Not started / In progress / Done |
| Define approved Bash commands and risky command deny rules | Engineering / DevOps | Not started / In progress / Done |
| Review MCP server and plugin marketplace policy | Platform / Security | Not started / In progress / Done |
Verify settings using /status and /permissions |
Pilot users | Not started / In progress / Done |
| Document user guidance and exception process | Change management / IT | Not started / In progress / Done |
| Review policy quarterly as Claude Code features evolve | Governance team | Not started / In progress / Done |
Vantage Point helps organizations adopt AI tools without losing control of data, workflows, or platform governance. For Claude Code and similar AI development tools, that means helping teams move from “people are using this” to “we know how this should be used safely.”
A practical engagement may include:
If your team is evaluating how Claude Code, Anthropic, Salesforce, HubSpot, integrations, or CRM governance fit together, Vantage Point can help assess the right next step and build a practical implementation plan.
Claude Code admin settings are being unified so Desktop reads the same managed admin policy as the CLI and IDE. Admins should manage Claude Code through server-managed settings, MDM policy, OS-level policy, or managed-settings.json rather than relying on retired Desktop-specific toggles.
June 5, 2026 is the date Anthropic identified for admins who want to explicitly disable auto mode or bypass permissions mode through organization Claude Code settings. If those modes are not already disabled, they are available by default according to Anthropic’s notice.
Businesses should disable Claude Code auto mode until they have a clear policy for when promptless execution is acceptable. Mature teams may later enable it for defined workflows with trusted infrastructure, narrow permissions, and monitoring.
Most businesses should disable bypass permissions mode on normal workstations because it skips permission prompts and safety checks. If it is used at all, it should be limited to isolated containers, VMs, or controlled development environments that cannot access sensitive systems.
Managed settings are organization-controlled policies that override user, project, and local Claude Code settings. They are used to enforce security requirements such as permission rules, sandboxing, MCP restrictions, plugin marketplace controls, and mode restrictions.
Admins can ask users to run /status in Claude Code to see which settings sources are active. Users can also run /permissions to review effective permission rules and confirm that managed rules are being applied.
This is a practical AI governance issue because Claude Code can act inside development and business operations environments. Organizations need policies for data access, file access, tool execution, integrations, review workflows, and exception handling before scaling autonomous AI work.
Yes. Vantage Point helps organizations connect AI governance to real operating workflows across Salesforce, HubSpot, integrations, data migration, and managed services. The goal is to make AI useful, secure, and supportable—not just available.