Cursor Cyber Safeguards & Anthropic CVP for Security Teams

Some legitimate security tasks get blocked, such as pentest planning, malware analysis write-ups and defensive tooling. Those blocks come from Anthropic's API safeguards, not from Cursor policy, and Anthropic applies them to its latest generations of Opus models. Cursor puts the CVP application and the model controls in its dashboard. Anthropic sets the program terms, decides on approval and sets the data handling for cyber-verified models.

That decides who you escalate to. A declined response on a malware write-up is Anthropic's API refusing, so a support ticket to Cursor will not move it.

Before you assume a block is a safeguard, run the same prompt on a model from another provider. A request that works elsewhere and fails on a recent Opus model fits the pattern. One that fails everywhere is more likely a problem with the prompt.

Set expectations on timing as well. CVP is an application that Anthropic reviews, and approval isn't instant. Cursor checks for approval every two hours, so plan near-term work around the models you can already use.

Cyber Safeguards apply at the Organization Group level (Organization → Groups), not to billing groups or SCIMSystem for Cross-domain Identity Management. A standard for automatically creating and removing user accounts when people join or leave. Press Enter for the full definition. directory groups. Only members of the approved group get cyber-verified model access, and the rest of the organization keeps normal Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. behavior.

Several questions in Cursor's docs for this feature are about finding the right group. If a group's settings show Spend Limit Overrides and Auto-RunThe Cursor Settings > Agents setting that decides which commands the agent runs automatically and which wait for your approval, via allow and block lists. Press Enter for the full definition. Controls but no Cyber Safeguards section, you are likely in team directory groups or billing groups. Organization Groups open from your profile menu in the bottom-left corner. A team's SCIMSystem for Cross-domain Identity Management. A standard for automatically creating and removing user accounts when people join or leave. Press Enter for the full definition. directory group, such as one synced from Okta, doesn't qualify either, because Cursor treats directory groups as a separate concept for team-level spend and policy. Links in the program email can also land you on the org Overview page, which doesn't hold these settings.

Keep the group small. The Cyber option grants access to that group's members and nobody else, so the right size is the smallest group that covers the people doing defensive work.

Take a five-person Security group with Cyber turned on for Opus 5. When one of them asks Opus 5 to analyze a malware sample, that request runs under Anthropic's CVP terms, without zero data retention. If the same conversation then uses a different model, those requests follow normal Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition.. Nothing changes for anyone outside the group.

Applying is a group-scoped request, not an org-wide switch. You set up a dedicated group, accept Anthropic's terms as an authorized representative, then wait for approval before you turn on the Cyber option.

1. 1Click your profile in the bottom-left corner, choose Organization, open Groups, and create or select a dedicated group (for example Security or CVP).
2. 2Add only the members who need cyber-verified model access.
3. 3In Cyber Safeguards Models Settings on the group's settings page, click Apply. Click I understand only if you agree to Anthropic's terms and are authorized to apply for your company, then click Save.
4. 4Wait for Anthropic's approval. Cursor rechecks every two hours and shows Approved once you're cleared.
5. 5Turn on the Cyber option for Opus 4.7, Opus 4.8 or Opus 5 in the same group's model access settings.

Read the terms before anyone clicks Apply. Confirming them states that you are an authorized representative of your company and that you agree to Anthropic's terms, so the person who clicks needs that authority. The resulting agreement is between your organization and Anthropic. Cursor is not a party to it, sets none of its terms and takes on none of its obligations. Only Organization admins can apply, because they create the group, so the work lands with whoever owns your org structure.

While you wait, Anthropic reviews the application and contacts you directly. The dashboard marks the group Approved on its own, so nothing needs resubmitting. Cursor says it isn't responsible for tracking the application, and the Anthropic portal shows its status in real time.

Put the privacy consequence in the change request, before anyone applies. The exception is narrow, covering the cyber-verified models for one group, but a reviewer would still rather approve it in advance than read about it afterward. Name the group and the models in the request, and record who signed it.

Cursor documents Cyber mode for Opus 4.7, Opus 4.8 and Opus 5. Pricing is the same as the base model, at the rates on Cursor's models and pricing page. In Cursor's steps the option comes after approval, so check that the group shows Approved before you look for it.

Mythos is not part of the program. If your security group usually works in another model, CVP approval changes nothing for that workflow. For the work that needs the safeguards lifted, they have to switch to an eligible Opus version, so plan for a change of habit as well as of permissions.

Base-model pricing removes the cost question from the business case and leaves the harder one. Someone has to own the decision that a named set of people works under different retention terms from everyone else. That is easier to get signed while the set is five people. Anthropic sets those terms, and Cursor's docs describe the loss of zero data retention as Anthropic's policy today, so give the approval a review date.