@Shivam_Singh2 — Adding my case here as it is the exact same issue: account-specific HTTP 500 on a Google Workspace account with an active AI Ultra subscription.
juergen_Meyer — I feel your pain. I’m on Day 36 now. Same scripted replies, same runaround.
My Case Details
- Support Case: #67960581
- Account: Google Workspace (AI Ultra subscriber)
- Issue Start Date: ~February 14, 2026
- Duration: Over 5 weeks and counting — second full month of paying for a completely unusable service
- Subscription Tier: AI Ultra (subscribed specifically for Antigravity agent access)
The Problem
Every prompt — including a simple “Hello” in a brand-new empty workspace — instantly fails with:
{
"error": {
"code": 500,
"message": "Internal error encountered.",
"status": "INTERNAL"
}
}
This happens on every model (Gemini 3.1, Gemini 3.0, Claude, GPT-OSS). The error occurs at the entitlement/authentication verification layer before any model is even loaded. This is NOT a capacity issue — it is a backend failure to validate my AI Ultra subscription state.
Trace IDs (for Engineering)
Latest reproduction (March 22, 2026 — today, fresh reinstall):
- Trajectory ID:
a69c3b76-2a73-4559-8eca-c8299ca1969f
- Sherlog TraceID:
0x3c69b8055dc7090e
- X-Cloudaicompanion-Trace-Id:
3c69b8055dc7090e
- Error: HTTP 500 Internal Server Error
Earlier reproduction (March 12, 2026):
- Trajectory ID:
b7acb2a6-3c6a-45ba-8b42-06deb00925c3
- Sherlog TraceID:
0xffd9b06471ea425e
- X-Cloudaicompanion-Trace-Id:
ffd9b06471ea425e
Original traces (February 14, 2026):
0x6c6064c9bb63ff570xd6e5ae26b4d2784a
What I’ve Already Done (Exhaustive Local Troubleshooting)
- Full uninstall and reinstall of Antigravity (done again today, March 22 — same error)
- Wiped all AppData/cache
- Tested on multiple networks
- Disabled all extensions and MCP servers
- Used “Reset onboarding” command
- Reset the license/entitlement from settings — did NOT fix the issue
- Tried all available models (Gemini 3.1 Pro High, Gemini 3.1 Pro Low, Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS 120B)
- None of these had any effect — consistent with a server-side root cause
The Support Experience (5+ Weeks of Canned Responses)
Here is the timeline of my interactions with Google Workspace Support — this is what “support” looks like for a paying Ultra customer:
- Feb 15 — Opened case via live chat. Agent Munjeti acknowledged the issue, confirmed it was affecting users globally, and promised to monitor personally.
- Feb 21 — Munjeti sends a template response about “capacity constraints” and suggests switching to Gemini 3.0. (Doesn’t work — the error is pre-model.)
- Feb 25, 27 — Automated “we haven’t heard from you” closure warnings.
- Feb 28 — Same template as Feb 21, word for word.
- Mar 4 — Another automated closure warning.
- Mar 5 — New agent (Malyala) sends the same “switch to Gemini 3.0” template.
- Mar 12 — Munjeti says “the issue has been resolved” by Product Engineering (capacity fix). It was NOT resolved for my account. I replied immediately with fresh trace IDs proving the error persists.
- Mar 13 — Munjeti acknowledges my reply, says the case is a “high-priority blocker,” Tier 3 and Engineering are engaged, and logs my billing credit request.
- Mar 18 — 5 days later, the same Munjeti sends a completely contradictory response telling me to “purchase the AI Ultra Access add-on.” I already have AI Ultra. This is the most telling evidence that no one is actually reading the case. It’s pure copy-paste.
- Mar 20 — Another automated closure warning.
About Claude and Model Usage
I want to clarify something important: Claude was available as a model inside Antigravity from the moment I installed it — it was included in my Ultra subscription from day one. I did not add it manually or do anything unusual. The models available in my Antigravity installation include: Gemini 3.1 Pro (High), Gemini 3.1 Pro (Low), Gemini 3 Flash, Claude Sonnet 4.6 (Thinking), Claude Opus 4.6 (Thinking), and GPT-OSS 120B (Medium).
I used Claude for code generation and Gemini for brainstorming/analysis as part of normal usage. I never exceeded any daily quota — in fact, my usage logs should show minimal activity because the agent crashes on the very first prompt every time. If my account was flagged or blocked because of using Claude, that is Google’s problem for making it available and then penalizing users who chose it. I did not misuse anything.
Key Points for Shivam_Singh2 and the Google AI Team
- This is account-specific, not general capacity. My Workspace account throws HTTP 500 while the service works for others. This matches sft’s exact observation.
- Claude was available from installation and I used it normally. I did not add or misuse any model. It was part of the Ultra subscription offering. If users are being blocked for using an officially available model, this needs to be acknowledged and fixed.
- Please check my actual usage logs. They will show minimal usage — nowhere near any quota limit — because the agent crashes on the very first prompt every time.
- I have been paying for AI Ultra for two months solely for Antigravity access. Zero value received. I am requesting a full billing credit or refund for February and March 2026.
- Support is actively counterproductive. Telling an Ultra subscriber to “purchase AI Ultra” is not just unhelpful — it’s insulting and proves the responses are automated templates with no case-level review.
- I reset the license/entitlement from Antigravity settings today — it did NOT fix it. This further confirms the corruption is server-side, not local.
Request to Google Engineering
Please investigate the entitlement state for my specific account. Based on the error pattern, the backend entitlement database appears to have a corrupted or stale record for my Google Workspace account. A forced resync/reset/reprovision of my AI Ultra entitlement tokens should resolve this.
If this thread gets attention from the engineering team, my Workspace Support case number is #67960581 — please cross-reference internally.
Thank you @sft for filing this thread with such clear technical detail. It perfectly describes what many of us are experiencing.
Visual Evidence (Screenshots from today, March 22, 2026)
Screenshot 2: Typing “Hello” results in immediate “Agent terminated due to error” — HTTP 500 on fresh install:
