I’m testing short Persian text-to-speech in Google AI Studio on macOS.
Manual Run in a normal signed-in Chrome tab has produced playable, downloadable audio. When a ChatGPT/Codex computer-use assistant activates Run, generation starts and then AI Studio reports:
“Http response at 400 or 500 level, error: , http status code: 403”
Observed:
- Assisted Run returned 403 in Chrome and Microsoft Edge.
- Native accessibility activation in Chrome also returned 403.
- Connecting the official ChatGPT browser extension in Edge allowed the assistant to select the Achird voice, edit the prompt, and activate Run, but generation still returned 403.
- A separate Playwright test with a clean temporary Edge profile was stopped before generation because Google rejected sign-in with “This browser or app may not be secure.” No workaround or bypass was attempted.
The test uses Gemini 3.1 Flash TTS Preview, Composer mode, one speech block, blank Sample Context, and the Achird voice. This concerns website generation, not a direct API call. No billing change or security setting modification was made.
Could you please clarify:
- Is assisted browser operation supported for the AI Studio speech-generation page?
- Is there a known reason manual Run succeeds while assisted activation returns 403?
- Is there a supported website-side remedy?
- What minimal diagnostic information can be shared securely?
The two failures may have different causes; I’m not claiming that automation detection caused the generation 403.