CISA Admin Leaked AWS GovCloud Keys on GitHub

· cloud devtools history · Source ↗

TLDR

  • A Nightwing contractor’s public “Private-CISA” GitHub repo exposed AWS GovCloud admin keys, plaintext passwords, and CISA internal system credentials for ~6 months.

Key Takeaways

  • Repo named “Private-CISA” contained importantAWStokens (three AWS GovCloud admin credentials), AWS-Workspace-Firefox-Passwords.csv with dozens of plaintext logins, and access to CISA’s Artifactory build pipeline.
  • The contractor manually disabled GitHub’s built-in secret scanning, then used the repo as a sync scratchpad between work and home machines since November 2025.
  • Exposed AWS keys remained valid for 48 hours after CISA was notified and the repo was taken down.
  • Artifactory access is the highest-severity vector: an attacker could backdoor software packages and propagate implants through every subsequent CISA build and deployment.
  • Many credentials followed the pattern platformname+year, weak even if never externally exposed; CISA is operating at roughly two-thirds normal staffing.

Hacker News Comment Review

  • Commenters flagged a compounding risk: LLM coding assistants reading .env files or secrets on disk can silently exfiltrate credentials to third-party model providers without triggering any secret-scanning alert.
  • Consensus is that the 6-month detection gap points to scanner coverage limits or growth outpacing tooling, not just individual negligence – with the broader takeaway being that external services like GitGuardian remain the last line of defense when internal controls are disabled.
  • The Artifactory supply-chain angle received little direct comment, but several users noted the staffing cuts as structural context explaining why no internal process caught this.

Notable Comments

  • @epistasis: LLM tools reading secrets from disk “ship it off to be training data” with no flags raised – a blind spot most secret-scanning workflows miss entirely.
  • @itintheory: Questions why scanners took months, not days, to surface the leak – suggesting GitHub’s scale may be outpacing automated detection pipelines.

Original | Discuss on HN