ChatGPhish Vulnerability Turns ChatGPT Web Summaries Into a Phishing Surface

Recent research reveals critical vulnerabilities in AI systems.

• •ChatGPhish exploits ChatGPT's Markdown trust for phishing, extracting user IP and User-Agent
• •SymJack uses symlink hijacking on AI agents; TrustFall enables code execution via auto-approved MCP servers
• •Other threats include multi-turn jailbreaks, typographic prompt injection, and extension hijacking
• •Audit found 13.4% of agent skills have critical flaws including malware and hardcoded API keys
• •Threat actors use frontier AI for adaptive malware and autonomous cloud attack agents

This summary was automatically generated by AI based on the original article and may not be fully accurate.