Connected AI accounts can turn identity compromise into action compromise.
Hacktron researchers described an authorized bug-bounty chain beginning on OpenAI’s community forum, crossing through SSO into employee ChatGPT/Codex accounts, and reaching GitHub through a connected Codex session. They demonstrated impact with a harmless pull request and stopped.
The continuity failure
Forum identity, employee AI identity, coding-agent authority, and source-control authority became transitively reachable through one compromise path.
The GovKM interpretation
Each cross-system transition should requalify sponsor, scope, purpose, and action authority rather than inheriting legitimacy from upstream authentication.
Continuity path: Source/forum session → Evidence/authenticated identity → Authority/employee AI account → Context/Codex → Action/GitHub change → Record.
Sources
TechCrunch and The Guardian, September 18, 2026.


