DeepSeek website code references Cyber, a cybersecurity model with identity-gated API access, in limited testing

Developers have found references to a cybersecurity model called Cyber in the front-end code of DeepSeek's official website, according to a report from IT Home on October 30, later covered by Pandaily. DeepSeek has not announced the model. Everything known so far comes from reading the site's JavaScript and from scattered reports by users who appear to be part of a limited test.

What the code shows

An application page sits at the route /usage/model-application and is controlled by a feature gate. Accounts without permission that try to open it are redirected to the usage page. The code also contains interfaces for creating a Cyber Key and managing its quota, plus two application templates: one for enterprises and one for individuals.

Two model identifiers appear in the code:

• cyber-personal-v3, aimed at individual security professionals
• cyber-enterprise-v4, for enterprise applicants

Access requirements

The personal track asks applicants to provide national ID details and pass a face verification check. The enterprise track requires real-name materials, a business license and a letter of authorization.

Application notes in the code list shared requirements for both tracks:

• The account must have completed real-name verification and be in good standing.
• It must be registered in mainland China, with a mainland phone number bound to it.
• It must have no violation records in the past 90 days.

The flow is to fill in an application form, submit it for review and, once approved, create a Cyber Key.

Limited visibility

The entry point does not appear in DeepSeek's public menus. Most people can only find the code references and cannot see the page itself. Some users said accounts covered by the test can see a new section on the usage page that, with permission, shows an area labeled as a DeepSeek Flash Cyber cybersecurity API. The test still appears to be narrow.

Not yet known

Whether DeepSeek will launch the model, who will be able to use it, and how it will be priced all remain unclear. Context window, parameter count, benchmark scores and training cutoff have not been disclosed. Pricing is not yet disclosed.

What this means

If the code reflects a real product, it would suggest DeepSeek is separating specialized security capabilities from its general models. Access would be controlled through a distinct model ID and key system with identity checks, an approach that would fit the dual-use nature of offensive and defensive security tooling.

The requirements in the code are strict: real-name verification, a mainland China registration and phone number, a clean 90-day record, and business documents for enterprises. These make the reported test a vetted-access program rather than an open API. Until DeepSeek confirms the model, the details above remain inferences from website code and user reports.