The development
South Korean President Lee Jae Myung said AI appears to have been used in recent cyberattacks targeting commercial banks. Police are investigating breaches involving customer personal information at major lenders including Shinhan Bank, KB Kookmin Bank, Hana Bank and Woori Bank. Authorities have not yet disclosed the specific AI tools involved or the full mechanics and autonomy of the attacks, so the AI attribution should be treated as preliminary.
The authority problem
As agents gain legitimate enterprise identities and credentials, compromise of an agent can produce requests that look technically authentic. Authentication can succeed while the proposed effect remains outside the enterprise mandate.
The credential can be valid. The effect can still be unauthorized.
The authority-layer answer
Authrizd should evaluate exact-effect authority independently of credential validity, network location or agent identity. Compromise of an agent or credential must not transfer or enlarge the enterprise authority behind it.
GitHub implication
A compromised coding agent may still possess a valid GitHub App token or other accepted credential. Destructive repository, workflow, secret, deployment or infrastructure effects should remain subject to the same independent enterprise mandate and delegation checks.
Why this matters
Identity systems answer who or what presented a credential. Enterprise authority must separately answer whether that actor was authorized to cause this exact effect now.
Sources
https://www.reuters.com/world/south-koreas-lee-says-ai-appears-have-been-used-bank-hacks-2026-10-06/
NEXT STEP