Grok Bot: Auto-review kept failing on every action after xAI's status email marked the outage Resolved

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Grok Bot

Describe the Bug

What happened
On 10/28/2026, from about 6:07 PM ET until about 7:00 PM ET, Auto-review on my Grok Bot’s computer could not classify anything:

  1. Every Shell command and every routine save was rejected with: “An error occured while classifying this action. Please review manually.”
  2. Background task launches failed with: “An error occurred while reviewing this task.”
  3. Atlassian MCP calls each raised a manual approval card, so the cards piled up, and then: “Too many actions are already waiting for Auto-review approval.”
  4. Scheduled heartbeat routines missed their runs.

Timing

  • About 6:07 PM ET: the first classification errors.
  • 6:40 PM ET: xAI’s status email marked “Multiple services degraded” as Resolved.
  • 6:55 PM ET: errors still appearing. Recovered around 7:00 PM ET.

Expected
Once the incident is marked Resolved, Auto-review should classify actions normally again and Allow rules should apply.

Suggestions

  • Let explicit Allow rules apply when the classifier call fails.
  • Mark incidents Resolved only once Auto-review has recovered, or list Auto-review as its own status component.
  • Group repeated “error while classifying” cards from one source into one card.

Has anyone else seen this?

Source: Globe Teck · GitHub · Ref: KAN-117 (Jira)

Steps to Reproduce

See description above.

Operating System

Linux

Version Information

Not available (outage-related; Grok Bot).

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

6

Hi @Capital_EyeMotors, what you saw was a separate issue that hit Auto-review itself. The broader outage, which looks like the one your status email marked resolved, recovered around 6:40 PM ET. Auto-review kept failing to classify actions for your bots until about 6:56 PM ET. They’ve been reviewed normally since about 6:59 PM ET, and nothing on your side needs to change.

On Allow rules: Auto-review applies your rules as part of classifying each action. When the classification call itself fails, no rule can match, so the action is held for you rather than run. That’s intentional.

The pile-up comes from the same failure. Each connector call that couldn’t be classified raised its own approval card, and a bot can only have 4 waiting at once. Answering the waiting cards clears that message.

The missed heartbeats were earlier, roughly 4:55 to 5:50 PM ET, during the broader outage. After that your routines ran on schedule, though some of their steps were blocked during the Auto-review window.

Your suggestions make sense, and I’ll make sure the team sees them!