🛡️ OpenAI–Anthropic cross-evaluations expose jailbreak risks and raise the bar for GPT-5 enterprise tests
2025-08-29OpenAI and Anthropic run cross-tests that surface jailbreak and misuse vulnerabilities, underscoring gaps that enterprises need to address in upcoming GPT-5 evaluations. Additions include rigorous adversarial testing, misuse resilience checks, and safeguards against policy circumvention and prompt injection.
Read more →