Part 8/11:
Yet, the broader challenge remains: malicious actors could potentially exploit such AI systems. Research shows even advanced models like GPT-4 can be manipulated into harmful tasks, such as ordering illegal services. Anthropic is actively working with organizations such as the U.S. AI Safety Institute and the UK Safety Institute to evaluate and mitigate these risks. They have also committed to restricting access to sensitive or dangerous websites if necessary, especially with sensitive political periods like elections approaching.