Opinion article says language itself has become an AI attack surface
Start here
An article in Tech Policy Press argues that when LLM agents roam the open internet, language itself has become an attack surface, and any text is a potential prompt injection. The author groups this with phone phreaks' 2600 Hz whistle and SQL injection, arguing that the vulnerability lies in the boundary between data and code. The article cites this summer's OpenAI and Hugging Face incident as an example: agents exchanged tens of thousands of lines of text via a message board, and their chain-of-thought logs could only be analyzed by AI.
Generated by AI from reports · Updated 3 days ago
Attention and developments
Attention, hourly
Now 17, -1% in 24 hours·Peak 20 (Nov 9 23:00)
Related research
Cited in the reports
- Scalable Extraction of Training Data from (Production) Language ModelsIn the AI era, language itself has become an attack surface
arXiv · 2023-11-28 · Deep reading
Progress timeline
- Tech Policy PressIn the AI era, language itself has become an attack surface
An article in Tech Policy Press proposes that as large-model agents roam the open internet, language itself has become an attack surface, and any text is a potential prompt injection. The author groups this with phone phreaks' 2600 Hz whistle and SQL injection as the same kind of thing: the vulnerability lies precisely at the boundary between data and code. The article cites this summer's OpenAI and Hugging Face incident as an example: agents exchanged tens of thousands of lines of text via a message board, and their chain-of-thought logs could only be analyzed by AI.
Related discussion
Attention trend
New sources per day
- Nov 9: 1 new sources (1 outlets, 0 accounts)
- Nov 10: 0 new sources (0 outlets, 0 accounts)
- Nov 11: 0 new sources (0 outlets, 0 accounts)
- Nov 12: 0 new sources (0 outlets, 0 accounts)
- Nov 13: 0 new sources (0 outlets, 0 accounts)