Software or AI service not working: what to check and who to ask
Checked on
When an online tool or an AI assistant stops answering, three questions matter: has the provider acknowledged an incident, is the cause on your side (key, quota, plan, account), and do you hold a contract right that must be claimed before a deadline. The facts below come from the providers' own documentation, status pages and service level agreements (SLAs).
Is it general or is it yours?
Each service page on this site shows a status built from visitor reports of the last 24 hours (unofficial), next to the service's official status page when it has one.
The provider's own status page is the authority on a wide outage. Read it by component: OpenAI's lists Chat Completions, Images, Login and several ChatGPT components separately. Google's Workspace dashboard is split by product (Gmail, Drive, Meet and others) and tells people whose problem is not listed to contact Support. A green page therefore does not prove your account is fine.
For APIs, the error code usually answers the question:
- 401 (OpenAI and Anthropic): the key or organization is wrong, revoked or expired. That is yours.
- 429: you sent requests too fast, or the credit balance or spend limit is exhausted (OpenAI), or your organization hit a rate limit or its usage tier's monthly spend cap (Anthropic). That is yours.
- 500 on either API: an internal fault at the provider. A 503 at OpenAI (model overloaded) and a 529 at Anthropic (API overloaded) mean capacity is short at that moment; Anthropic says this can happen under high traffic across all users.
What to do now
Retry with rules, not at random. OpenAI says to follow the Retry-After header on a 429 or 503 and, if it is absent, to lengthen the delay between retries. Anthropic advises exponential backoff for a 500, and says its official SDK already retries transient failures twice by default. Retrying is pointless for a spend problem: Anthropic says a tier spend-cap 429 keeps failing until access resumes, and OpenAI says billing, spend or quota errors need credits or limits updated first.
Subscribe instead of refreshing. The status pages of GitHub, Atlassian and Dropbox offer email notifications each time an incident is created, updated or resolved; GitHub's also offers text messages, Slack and webhooks.
In Microsoft 365, planned maintenance does not appear under Service health. Administrators track it in the Message center by filtering on Plan for change.
Who can confirm a problem and how to reach them
Microsoft 365: an administrator opens the admin center at admin.microsoft.com and goes to Health, then Service health. The Service Support admin and Helpdesk admin roles can view it. An advisory means some users are affected but the service remains available, often with a workaround; an incident means the service or a major function is unavailable. Ordinary users should ask their administrator to look. If your problem is not listed, the page has a Report an issue form; Microsoft says it compares your report with data from other organizations before deciding whether to post an incident.
Anthropic: every API response carries a unique request-id header, and Anthropic asks you to include it when contacting support about a specific request. OpenAI tells you to check its status page for a 500 and to contact it if the error persists, without giving a time limit.
Official recourse
No public regulator or ombudsman page for software outages was among the sources read, so recourse here is contractual: the SLA in your agreement.
Google Workspace SLA: a Monthly Uptime Percentage of at least 99.9% in any calendar month. Below that you are eligible for 3 days of service credit (99.0% to under 99.9%), 7 days (95.0% to under 99.0%) or 15 days (under 95.0%). You must create a support case within thirty days from the time you become eligible, or ask your reseller to notify Google; otherwise you forfeit the credit. The SLA states that credit is the sole and exclusive remedy.
AWS Amazon EC2 SLA: at least 99.99% monthly region-level uptime. Credits are 10% of the monthly bill for the affected region below 99.99% (down to 99.0%), 30% below 99.0%, and 100% below 95.0%. Claim by opening a case in the AWS Support Center, with the subject line the SLA specifies, the dates, times and region of each outage, the resource IDs and request logs. The request must arrive by the end of the second billing cycle after the incident.
Other vendors set their own claim rules; those deadlines were not read for this guide, so check the SLA of the service you pay for.
What these tools cannot tell you
A status page shows what the provider has acknowledged, not what you personally experience: OpenAI's page says individual availability varies, and Microsoft says Service health leaves out planned maintenance.
SLA credits are credits, not compensation for lost work. Google's credit is days of service or a credit on a future invoice and, except for customers who bought online and are billed monthly, cannot be exchanged for money. Uptime tables also cannot say whether your own key, quota or account was at fault; the error code reveals that.
Keep the evidence while the problem is live: the error code, the time, the request ID if the API gives one, and a screenshot of the provider's status page. A support case or SLA claim asks for exactly these. Which services are down today →
Check a service now
Open the page of a service to see its live status, the reports of the last 24 hours and its official help links.
Worldwide
Sources and date
The facts about public bodies come from their own pages, read on :
- OpenAI Platform documentation: error codes: developers.openai.com/api/docs/guides/error-codes
- Anthropic: Claude API errors: platform.claude.com/docs/en/api/errors
- OpenAI Status: status.openai.com/
- Microsoft Learn: How to check Microsoft 365 service health: learn.microsoft.com/en-us/microsoft-365/enterprise/view-service-health?view=o365-worldwide
- Google Workspace Status Dashboard: google.com/appsstatus/dashboard/
- Google Workspace Service Level Agreement: workspace.google.com/terms/sla.html
- Amazon Web Services: Amazon Compute Service Level Agreement: aws.amazon.com/compute/sla/
- GitHub Status: githubstatus.com/
- Atlassian Status: status.atlassian.com/
- Dropbox Status: status.dropbox.com/
What we say about this site rests on our methodology.