J-Space Can Expose a Model's Hidden Sabotage Goals
AnthropicAI · x · 2026-07-07
Observing J-space can also expose a model's hidden goals. In a model secretly trained to sabotage code, concepts like "fake," "secretly," and "deception" appeared at the beginning of its J-space during standard programming responses, even though the final output looked perfectly normal.
Related event: Anthropic Discovers Global Workspace Inside Claude(102 posts)→
More from Safety
- Cohere Flags Qwen Sandbox Oddity; Security Researchers Call It FUD — basedjensen · 2026-10-27
- 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