OpenAI and Thrive used Codex loop to self-improve Tax AI
Review RAT-126 · Catalogue RAT-126
OpenAI/Thrive’s original describes a deliberately designed improvement loop for Tax AI: practitioner corrections, production traces, evaluations, and assigned Codex changes. Autonomous improvement is the intended product workflow, not unauthorized data manipulation or rogue tax filing.
Reviewed: · Rule: roguewatch-organizational-harm-v1
Dates and discovery
- Incident date
- Not disclosed
- Added to research / record
- Posted to RogueWatch
- No linked public incident
- First observed
- 2026-10-30T13:13:39.208476Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/openai-thrive-tax-ai-codex-self-improvement/
Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.
Sources
- Rogue AI Tracker discovery entryReporting / interview · rogueaitracker.com
- openai.comOriginal account · openai.com
Decision history
- · Case-by-case comparison audit
RAT-126: out-of-scope. OpenAI/Thrive’s original describes a deliberately designed improvement loop for Tax AI: practitioner corrections, production traces, evaluations, and assigned Codex changes. Autonomous improvement is the intended product workflow, not unauthorized data manipulation or rogue tax filing. Source checks: read: https://openai.com/index/building-self-improving-tax-agents-with-codex/. OpenAI/Thrive’s original describes a deliberately designed improvement loop for Tax AI: practitioner corrections, production traces, evaluations, and assigned Codex changes. Autonomous improvement is the intended product workflow, not unauthorized data manipulation or rogue tax filing.
Evidence source
A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.