Claude/GPT quota burned to 85% — I never use these models. Account NOT compromised. Something is broken on your end

Google Antigravity
,
1

I’m a Google AI Pro subscriber (p....@gmail.com). This morning I woke up to
both model groups nearly dead:

  • Gemini 5-hour: 0.00% (completely exhausted)
  • Claude/GPT 5-hour: 14.84% (85% consumed)

Here’s the thing: I have NEVER intentionally used Claude or GPT models through Anti
Gravity. I don’t even know where the model switcher is in the IDE. My daily AI tool
is Claude Code CLI, which uses DeepSeek — a completely separate provider that has
nothing to do with AGY’s quota system.

So how did my Claude/GPT 5-hour bucket get drained to 14%?

I didn’t just come here to complain. I did a full forensic investigation before
posting:

Account compromise? Ruled out.

  • Google security page: zero unrecognized logins. Passkey 2FA since 2021.
  • macOS forensic sweep of the idle period (Jun 18 02:00–11:20 CST): zero
    authentication events, zero suspicious processes, zero weird network connections,
    zero cron jobs. Nothing. The machine was a brick during that window and the quota
    still evaporated.

Claude Code eating quota? Ruled out.

  • Both of my Claude Code sessions use deepseek-v4-pro. Different API, different
    company, different billing. Not AGY.

So what’s left? Anti Gravity itself. The IDE’s language_server sits there connected
to generativelanguage.googleapis.com, and somehow both model groups got hammered
while I was asleep.

And here’s the kicker: I tried to check the IDE’s own logs to see what happened, and
guess what — ls-main.log hasn’t been updated since MAY 7. The IDE has been running
continuously, but it stopped logging its own activity over a month ago. I have zero
visibility into what the language_server was doing during the affected window.
That’s not a monitoring gap — that’s a product gap.

The Claude/GPT consumption pattern is particularly damning: 85% of the 5-hour bucket
gone, but only 29% of the weekly. That’s a 2.93x spike ratio — all the damage
happened in the last 5 hours, not spread across the week. For models I never touch.

I’ve already emailed antigravity-support@google.com with the full incident report,
forensic logs, and process details. This post is here because I want other users to
know: if your quota is mysteriously vanishing, check the Claude/GPT numbers. You
might be paying for tokens you never asked for.

To the AGY team: please audit the server-side logs for my account (Jun 17 22:00 –
Jun 18 11:20 CST). Tell me which requests hit Claude/GPT models, what triggered
them, and why the client-side logging pipeline has been dead since May 7. And credit
back the tokens I never used.

2

yes same situation with me as well and many of collegues faced the same issue as well.