I’m a ChatGPT Pro ($200/month) user, and I’m posting this because a single long-running Astra Work task consumed almost half of my weekly allowance but ultimately gave me no usable final deliverable due to an OpenAI guardrail.
I have already opened a Support case and it has been escalated to a support specialist. I’m posting here as well in case OpenAI staff can help flag or investigate this failure mode.
The main issue is not simply that the task used a large amount of allowance. The issue is that the allowance was consumed, the work substantially completed, and then the result was withheld by the platform itself.
On September 21, 2026 (Asia/Taipei, UTC+8), an Astra Work task in one conversation ran for:
4h 16m 46s
During/around this run, my remaining weekly Astra allowance went from approximately:
68% remaining → 22% remaining
So approximately 46 percentage points of my weekly allowance were consumed.
The task had apparently already completed substantial work. The final Work summary reported completed research, repository edits, successful tests, and a generated PDF/report.
However, when the task reached final delivery, ChatGPT displayed:
“This content can’t be shown. Daybreak isn’t available for Astra. Some cybersecurity requests may still be limited.”
The result therefore became unusable/inaccessible even though the Astra computation had already consumed approximately 46% of my weekly allowance.
I am requesting restoration/reconciliation of the weekly allowance consumed by this failed run.
From a user perspective, this is materially different from choosing to spend 46% on a successful long-running task: the platform allowed Astra to run for more than four hours and consume the allowance, but an OpenAI guardrail prevented delivery only at the end.
There may also be a broader metering issue. Earlier in the same conversation, on September 20 around 20:30–22:00 Asia/Taipei, I observed approximately 12 percentage points of weekly allowance disappear, including one non-Fast Astra run of only around 10–20 minutes that appeared to consume ~7 percentage points.
However, even if those earlier charges turn out to be expected, I believe the 4h16m run needs separate review because the paid/limited resource was consumed while the corresponding final result was blocked by the service itself.
I have asked Support to inspect the server-side per-run metering history and determine exactly how much allowance was charged to the failed execution.
What I’m asking OpenAI to do:
- Review the server-side usage record for the 4h16m46s Astra Work run.
- Confirm the exact amount of weekly allowance it consumed.
- Restore/reconcile the allowance attributable to that run because the final deliverable was blocked by the Daybreak/Astra guardrail.
- Separately check whether the unusually fast usage earlier in the same thread reflects expected Astra metering or another accounting issue.
I can provide screenshots showing the 68% → 22% change, the 4h16m46s runtime, the completed-work summary, and the final Daybreak/Astra error.
I am deliberately not posting the private Work thread identifier publicly, but I have already supplied it to OpenAI Support.
If an OpenAI staff member sees this, I would appreciate having this linked to the appropriate Astra/Work usage or guardrail team.