Posted in AI Security
The first thing attackers spent after the Northflank breach wasn't your database. It was your AI API keys…
Craig M. — Chief Product Officer @ Cyberr®
The first thing attackers spent after the Northflank breach wasn't your database. It was your AI API keys.
On 30 October the cloud platform told customers to assume every environment variable had been decrypted and to rotate all keys. Users soon reported their LLM keys being used without permission.
That ordering is the lesson. A stolen AI key is cash on day one: no lateral movement, no exfiltration, just someone else's GPU bill on your account.
Yet most teams still treat those keys as dev config, not production secrets:
→ No spend caps or usage alerts on the provider side
→ One key shared across environments and services
→ Stored as plain env vars on a platform you don't control
If a key can spend money, scope it, cap it and monitor it like a payment credential.
Because to an attacker, that's exactly what it is.
6 reactions · 1 comments
Comments
Schomer: Excellent breakdown, Craig. This is a textbook example of a structural policy failure that traditional downstream security tools completely miss. The core issue here is that from a network and application standpoint, the traffic looks perfectly valid—the stolen API key authenticates normally, and the commands are well-formed. It’s an implicit trust failure. Attackers aren’t looking for lateral mov