←
Len Noe1,644 posts
Len Noe avatar

Len Noe

@hacker_213

Technical Evangelist/Whitehat/Transhuman

◷ Joined June 2018▧ 245 Photos and videos
Len Noe avatar
Len Noe@hacker_213·8h

Whoever can write to that Project JavaScript field runs code as your origin in every visitor's browser. That is privilege. Closer to prod deploy rights than a dashboard login. Nobody put it in the access review.

Len Noe avatar
Len Noe@hacker_213·8h

Report URI's Scott Helme found malicious JavaScript hitting at least 178 organizations. Per his numbers: 68 had no CSP, 50 had a policy that still let it run, 24 had one strict enough to block it.

Len Noe avatar
Len Noe@hacker_213·8h

Attackers did not hack the BBC, CNN or Zoom. They got write access to a field in an A/B testing tool. Those sites loaded the result from a domain they already trusted. Optimizely, November 16.

Post attachment
Len Noe avatar
Len Noe@hacker_213·Nov 16

Fix list: find every system that holds keys for other systems. One key per app. Narrow scope. Hard spend cap. Rotate in hours, not quarters.

Len Noe avatar
Len Noe@hacker_213·Nov 16

Amazon also says none of the techniques were new. Default creds, keys in code, no-login interfaces. The agents just did not get tired. Amazon cannot verify the outcomes they reported.

Len Noe avatar
Len Noe@hacker_213·Nov 16

Amazon's decoy network caught the agents' own notes between Oct 3 and Nov 5. Targets: LLM gateways like LiteLLM and One API. One weak admin page, many provider keys.

Len Noe avatar
Len Noe@hacker_213·Nov 16

Amazon says attacker AI agents used stolen model API keys to pay for the inference that hunted for more stolen keys. The attack funds itself out of your account.