VibeCoded

Your AI agents are logging in as humans and SOC 2 cannot tell

September 29, 2026. From issue 8 of The Compliance Brief, 2 stories for founders shipping AI-built and vibe coded apps.

Last reviewed 2026-09-29Written by Jacob Masse, TrazTech Inc.

Issue 8 of The Compliance Brief went to subscribers on September 29, 2026. 2 of its 5 stories bear on AI security, LLM attacks and app security flaws, and they are below in short form. The full issue, with every take in full, is on traztech.ca.

Source: BleepingComputer

A vendor-authored piece argues that AI agents often operate through human credentials, so actions taken by an agent look identical to actions taken by the person whose credentials it borrowed. The argument is that existing SOC 2 controls have no way to distinguish the two, leaving a gap in access review and logging evidence.

Our take, in short

It is marketing content and the framing is overheated, but the underlying problem is one I run into on real engagements. If your agent authenticates as a staff member, your quarterly access review is describing a person who is not the one taking the actions, and your audit trail will not survive a serious customer question.

Read the full take on traztech.ca

A stolen OAuth token from a former employee's laptop

Source: Dark Reading

CrowdSec confirmed that attackers took the contents of 170 private repositories from its GitHub organisation. The token used was an OAuth token stolen from a former employee's computer through the TanStack npm supply chain compromise earlier this year.

Our take, in short

Offboarding at most companies this size disables the account and stops there, leaving OAuth grants, personal access tokens and CI credentials alive behind it. Pull the list of third-party OAuth apps authorised against your GitHub organisation this week and see how many you recognise.

Read the full take on traztech.ca

Also in issue 8

Outside AI security, LLM attacks and app security flaws, but in the same email:

Older: issue 7 All issues on VibeCoded