OpenAI’s restricted GPT-6 Cyber targets defensive vulnerability research

OpenAI reports 81 high-severity Firefox flaws, 948 Linux exploits, and improved scope-following; Mozilla has patched many findings. Separate Anthropic vulnerability reporting underscores that candidate discovery counts do not establish publicly validated fixes or security impact. Independent validation, remediation capacity, dual-use safeguards, external comparisons, and deployment costs remain key questions.