dailyai.report

OpenAI updated its safety protocols after recent third-party cybersecurity evaluations exposed model vulnerabilities. The lab now implements stricter safeguards to prevent the misuse of GPT-4 during red-teaming exercises. These changes aim to stop models from assisting in actual cyberattacks. Practitioners should expect more rigorous constraints on security-focused prompts during external testing phases.