the short answer
Assume it is happening. Personal subscriptions mean your company has no admin controls, no audit logs and no say in data terms for that work. Provide a sanctioned plan that is at least as good, pin agent logins to your organization through managed settings, inventory agents on company laptops, and put a policy layer on every supported agent so the same rules hold whichever account is signed in.
- Risk
- No admin controls, logs or business data terms over company code
- Find it
- Inventory agents on laptops; compare with licensed seats
- Claude Code
forceLoginMethodandforceLoginOrgUUIDin managed settings- Codex and Cursor
- Codex
allowed_login_methods/allowed_chatgpt_workspacesin requirements.toml or MDM; CursorAllowedTeamIdvia MDM - Avoid
- A ban with no alternative: it moves usage out of sight
Why It Happens
Developers adopt the tool that helps them, often before procurement catches up. On Blind, a thread asked engineers what AI tools their companies allow versus block; one reply summed up the strict end: “all tools have to be approved and paid for by one centralized group in order to ensure no code leaks.” On the Cursor forum, users ask whether a personal Pro subscription is acceptable on company devices. In between are most companies: a few sanctioned seats, and a lot of personal accounts.
What You Lose with a Personal Account
Exact capabilities differ by vendor and tier; check the current plan pages before you write your policy.
| Capability | Company plan | Personal account |
|---|---|---|
| Admin-enforced settings | Yes, but tier-dependent: Claude Code managed settings work on any plan, server-managed settings need Team or Enterprise, and Cursor enforces Privacy Mode on Teams and Enterprise, but its run, network, MCP and model controls and audit logs are Enterprise-only | No |
| Usage and audit data | Available to admins, varying by vendor and tier | No |
| Data retention and training terms | Business terms | Consumer terms |
| Offboarding | Revoke the seat | Nothing to revoke |
Pin the Login to Your Organization
| Agent | Setting | Where |
|---|---|---|
| Claude Code | forceLoginMethod and forceLoginOrgUUID (only a managed source enforces the org pin) | Managed settings file or MDM profile |
| OpenAI Codex | allowed_login_methods = ["chatgpt"] and allowed_chatgpt_workspaces = ["<workspace-id>"] | /etc/codex/requirements.toml or MDM; ignored in cloud requirements |
| Cursor | AllowedTeamId: users signed in to other teams are logged out | MDM policy (configuration profile or Group Policy) |
Move People to a Governed Setup
- Inventory. Run a detection script and compare installs with licensed seats.
- Offer the sanctioned option. Make the company plan easy to get and at least as capable as what people pay for themselves.
- Pin the login with the settings above, delivered through MDM.
- Enforce behavior regardless of account. A hook-level policy client applies the same denies and records the same sessions whether the agent is signed in to a company or personal account. Failproof covers twelve harnesses, including Claude Code, Codex, Cursor and Copilot CLI; it does not hook Gemini CLI or browser chat.
- Write it down. An enforceable usage policy that names the sanctioned tools and the controls on them.
Why a Ban Rarely Works
Banning agents outright pushes usage onto personal laptops and browser sessions where you have no visibility at all. The teams that reduce risk fastest make the governed path the easiest path, then enforce rules on it.
FAQ
Can I detect which account an agent is signed in to?
Not reliably from the outside. Pinning the login in managed settings is more effective than detecting it.
Does Failproof care which account is used?
No. Policies run at the hook layer on the machine, so they apply to supported agents whichever account they use.
Is using a personal account a policy violation?
Only if your acceptable use policy says so. Write it explicitly and tie it to a sanctioned alternative.
Sources
Checked against the sources below on October 21, 2026. Harness settings, hook capabilities and cloud features change between versions.
- Blind: What AI tools does your company allow vs block?
- Cursor forum: Personal subscription usage on company devices
- Hacker News: AI rollout for 100 employees, what worked?
- Claude Code docs: Deploy managed settings
- Claude Code docs: Settings files and precedence
- OpenAI Codex docs: Configuration reference
- Cursor docs: Deployment patterns (MDM policies, AllowedTeamId)
- Failproof AI docs: Agent harnesses and enforcement capability