TAKO Okta AI agent lets strangers create the admin account before setup
Anyone on the network can claim the administrator account of this Okta helper tool before the real owner finishes setup. The attacker then controls the tool and the Okta data it reads.
| Severity | CriticalCVSS 9.8 of 10 |
|---|---|
| Being exploited? | UnclearNo reports either way yet |
| Fix available? | Yes3.0.1-beta · since 30 Oct 2026 |
| How sure is this? | Official sourceBacked by an official source |
How the attack works simplified
- 1Attacker scans networkFinds exposed port 8001
- 2Calls setup endpointPOST /api/auth/setup with no login
- 3Admin account createdSession cookie issued to attacker
- 4Okta data accessedReads users, tokens, configuration
- ResultAttacker controls the admin account
What happened
In TAKO AI Agent for Okta before 3.0.1-beta, the unauthenticated POST /api/auth/setup endpoint creates the first ADMIN user and issues a valid session cookie, guarded only by an is_setup flag. The default Docker setup publishes port 8001 on all interfaces.
Are you affected?
You are affected if you run TAKO AI Agent for Okta before 3.0.1-beta, especially with the default docker-compose.yml, and complete initial setup after the service is reachable by other hosts.
TAKO AI Agent for Okta, all versions prior to 3.0.1-beta
What to do
- Update TAKO AI Agent for Okta to 3.0.1-beta or later
- If you cannot update yet, do not expose port 8001 to any network and complete setup immediately on a trusted host
- Check whether an unexpected admin account exists and rotate Okta API tokens and integration credentials if you find one
Timeline
- 30 Oct 2026, 23:17 UTCPublicly disclosed
- 30 Oct 2026, 23:17 UTCReported by NVD · NVD
- 30 Oct 2026, 23:20 UTCPicked up by Zero Radar
Sources 1 report, oldest first
Official vendor, government CERT or advisory database
- CVE-2026-63377NVD · 30 Oct 2026, 23:17 UTC
Technical details
- ID
- CVE-2026-63377
- CVSS
- 9.8 / 10
- Disclosed
- 30 Oct 2026, 23:17 UTC
- On Zero Radar
- 30 Oct 2026, 23:20 UTC (3 min after disclosure)
- Last change
- 30 Oct 2026, 23:20 UTC
The plain-language parts and the attack diagram were written by an AI model (anthropic/claude-haiku-5.5) from the sources above. Facts like versions, exploitation and dates come from the sources themselves. How Zero Radar works