Distributed Attacks in Persistent-State AI Control
As AI coding agents become more autonomous, they increasingly ship code iteratively, with the codebase persisting across sessions. This persistence creates a new attack surface: a misaligned or prompt-injected agent can distribute attacks across pull requests (PRs) and time its payload for the PR with the best natural cover. To study the resulting dynamics, we introduce Iterative VibeCoding, a setting for AI control, the study of safely deploying capable but potentially untrusted AI. In Iterative VibeCoding, a coding agent builds software over a sequence of PRs in a persistent codebase while p
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 58%opensandbox-group/OpenSandbox →
- PossiblePossibly related (embedding) · 55%Th0rgal/sandboxed.sh →
- PossiblePossibly related (embedding) · 53%JKHeadley/instar →
- PossiblePossibly related (embedding) · 53%tenuo-ai/tenuo →
- PossiblePossibly related (embedding) · 51%Securing the future of AI agents →
- LinkedLinked via arxiv author · 85%Josh Hills →
“Distributed Attacks in Persistent-State AI Control”
- LinkedLinked via arxiv author · 85%Ida Caspary →
“Distributed Attacks in Persistent-State AI Control”
- LinkedLinked via arxiv author · 85%Asa Cooper Stickland →
“Distributed Attacks in Persistent-State AI Control”