Hacker News AIT2·85 pts
- Attackers exploited ChatGPT's standard sharing feature to inject malicious code without requiring users to install plugins or enable developer tools.
- Malicious scripts executed under the chatgpt.com domain with full access to the victim's authenticated session, enabling direct internal API calls.
- Key lesson: Any public or shareable surface must be treated as attacker-controlled input, and cross-frame messages require strict authentication.