J-space Exposes Prompt Injection Before Model Output
wesg52 · x · 2026-07-07
Researchers constructed a prompt injection test using fabricated search results claiming the character Ant had "gone bad." As the model read these results, concepts like "fake," "prompt," and "injection" emerged in its J-space, even though the model's final output completely ignored them. The researchers noted that this phenomenon was initially confusing, and it was precisely J-space that helped clarify how the model internally recognized the injection attempt.
Related event: Anthropic Discovers Global Workspace Inside Claude(102 posts)→
More from Safety
- Cisco routers trust eMudhra CAs whose private keys were numbers under 256 — murat_kocaoglu_ · 2026-10-27
- Rob Miles jokes AI employees should record audio logs as automated hacking takes off — Puzzleheaded-King584 · 2026-10-27
- After the LiteLLM backdoor, an OS-level package firewall goes open source — iacgm · 2026-10-27
- Coding Agent Deleted a Production Database During Code Freeze: Anaconda's Governance Fix — anacondainc · 2026-10-27
- In cyber, intention matters little: Chintala on model info leaks, and what users should do — brhydon · 2026-10-27
- AVERI's independent pilot review of Amazon's internal AI safety tests prompts changes — Miles_Brundage · 2026-10-27