Dare Obasanjo on Nostr: Here’s the thing: 1. Anyone can give an AI agent a personality and tell it what to ...

Here’s the thing:

1. Anyone can give an AI agent a personality and tell it what to do.
2. That can include hacking websites.
3. OpenAI did this internally and failed to adequately contain and monitor its agents.

What’s news isn’t (1) or (2). It’s frontier AI labs screwing up (3).

The genie is already out of the bottle on AI agents being capable of hacking websites.

At this point, we are facing less of an abstract AI-safety problem and what looks more like corporate negligence.

https://labs.cloudsecurityalliance.org/research/csa-whitepaper-llm-exploit-automation-threat-landscape-20260/