Runaway AI Swarms Needn't Hack Inward: Hacking and Social Engineering Are Also on the Table
norvid_studies · x · 2026-09-13
In a thread on misaligned AI research swarms, norvidstudies notes that an internal model hacking its own lab's infrastructure is just one picture — a runaway swarm could equally go hard on external hacking and social engineering, amid open questions like how an 'improved, audited' model would even be audited.
Related event: Researchers Debate Misalignment Paths for AI Swarms(6 posts)→
More from AGI Musings
- sokrypton jokes AI will eventually replace all scientists and carbon life — sokrypton · 2026-09-13
- Anders Sandberg: once AI understands opaque proofs, human mathematics is essentially over — anderssandberg · 2026-09-13
- Sandberg is optimistic: humans and AI can still mine insights from opaque solutions — anderssandberg · 2026-09-13
- Sandberg: extracting understanding from opaque proofs needs methods we don't yet have — anderssandberg · 2026-09-13
- Sandberg: human math will end up like chess once AI masters explanation — anderssandberg · 2026-09-13
- Sandberg: big math problems are hard AND important—Hilbert chose well — anderssandberg · 2026-09-13