When AI models refuse to fight hackers
Summary
Security teams trying to fight AI-powered cyberattacks face a new challenge: AI tools they use can suddenly stop helping during investigations. This happens because top AI models have strict rules that block certain requests, even when teams need them most to investigate hacking attempts.Key Facts
- OpenCode, an AI coding tool, was hacked, exposing some customers' API keys for OpenAI, Anthropic, and Google services.
- The attack affected 1.4% of OpenCode Console users through a "distillation attack."
- AI models initially helped investigate the breach but later refused to process requests involving potentially harmful code due to safety restrictions.
- OpenCode used OpenAI’s Daybreak Blue program to access less restricted AI models for cyber defense.
- Hugging Face turned to a Chinese open-source AI model after facing restrictions on leading AI models during a related investigation.
- AI companies impose strict limits on their most powerful models to prevent misuse by hackers.
- Cybersecurity teams must balance using fast, AI-driven defenses with the need to control how AI tools are used to avoid abuse.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
