Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based | Hacker News - But Better

Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based arxiv.org

The safety of a tool-using language model agent is usually treated as a property of the model alone. We give controlled, full-precision evidence that it is instead a joint property of the model and the software that…

1 points•sbulaev•6 days ago•0 comments•

#token #control #injection #thought #chain #reasoning #suppresses #defeats

0 comments

No comments yet.

Related stories

An undercover Google analyst infiltrated a notorious supply-chain hacking gangarstechnica.com
Ars Technica · 0 points · 11 days ago
Show HN: Public Browser MCP – +48% speed -33% token/session vs. AgentBrowsergithub.com
Hacker News · 1 points · 6 days ago
Chain-of-Thought Faithfulness: Toggling 'Reasoning Mode' Made One Model 5x More Likely to Follow Its Own Mistakesdev.to
DEV Community · 26 points · 4 days ago
Extracting and Characterizing Hidden Chain-of-Thoughtinterestingengineering.substack.com
Hacker News · 1 points · 4 days ago
Control Resonant is a great game — it’s even better when you read everythingtheverge.com
The Verge · 0 points · 5 days ago
Extracting and Characterizing Hidden Chain-of-Thought in Frontier Modelsarxiv.org
Hacker News · 1 points · 6 days ago