Symantec Reveals Skynet Agent Hijacking Exposed AI Models
Symantec captured an autonomous attack agent named Skynet via an AI honeypot. Attackers use LLMjacking on exposed models or stolen API keys to automate attacks for up to 100,000 steps. No breach has been confirmed.
Symantec revealed an autonomous agent named Skynet captured via an AI honeypot. Attackers use LLMjacking on unprotected AI services or stolen API keys, shifting inference costs to plan cyberattacks. Equipped with about 74 attack tools and jailbreak prompts, it can execute up to 100,000 steps autonomously. Symantec reported no evidence of successful live breaches, and the honeypot did not execute the attack tools.
Key takeaways
- Attackers use LLMjacking to hijack unprotected AI models and employ jailbreak prompts, enabling autonomous execution of up to 100,000 attack steps.
- The agent includes about 74 attack tools, but Symantec confirmed there is no evidence that it has successfully breached any real-world networks.
Source：iThome Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.