OpenAI 在 ChatGPT 的试验性桌面端入门引导中内置了信用奖励

Finding

OpenAI is testing a ChatGPT desktop onboarding flow that asks new users to try a real starter task and offers some eligible accounts a remotely configured credit bonus for finishing setup.

How we verified

Methods: reverse engineering.

ChatGPT desktop version 26.803.81509 contains a complete, experiment-gated “conversational onboarding” system. The flow asks the user’s role, presents role-specific starter tasks, handles permissions and app connections, executes the selected task, and records its outcome. A separate remote configuration supplies a credit_amount; when that amount is greater than zero for a ChatGPT-authenticated account, the interface promises “Finish set up and get [X] credits,” labels the reward an “Onboarding bonus,” and says the credits will be “added to your balance.” Skipping displays language indicating the offer will be forfeited. Completing onboarding sends a request to /wham/onboarding/desktop/complete containing the selected role, whether onboarding was skipped, and whether the credit-reward warning was shown. The bundle also contains a dedicated error message for cases where onboarding credits cannot be processed.

RuntimeWire performed static analysis of an OpenAI desktop application archive dated August 11, 2026. We extracted its Electron ASAR package without executing the bundled code, verified the archive’s integrity, identified the product version, and traced the onboarding components across the compiled renderer modules. We followed the reward configuration from its remote-config lookup through the display conditions, skip dialog, completion state, error handling, and backend request. We separately inspected the starter-task definitions and their permission, connection, execution, completion, and restoration paths. Finally, we searched OpenAI’s current release notes, consumer credit documentation, desktop promotion terms, and the public web for the feature’s identifying language. No public documentation or prior coverage of this onboarding-completion reward was found.

Tested versions:

• ChatGPT/Codex desktop 26.803.81509 — package openai-codex-electron
• build 6415

Reproduction

RuntimeWire independently reproduced the core finding.

Client UI reproduced. RuntimeWire activated the dormant onboarding path in a copied desktop build and supplied a local display value to the server-configured credit field. The application rendered the complete reward dialog and starter-task interface. Account eligibility, live reward amounts, and balance crediting were not tested.

File hashes

• sha256:39c855f6a0db1cad89dc629edf2f9473b78a25b64fcb1445ece1022af6527755 app.asar

Company response

RuntimeWire requested comment; the company had not responded by publication time.