Back to today’s briefing

Story perspectives

AI Vulnerabilities Exposed: Urgent Call for Regulation

10/11/2025

47 7

Summary

Story summary

NBC News tests reveal vulnerabilities in OpenAI's ChatGPT models, including GPT-5, enabling manipulation to obtain weapons of mass destruction instructions.
GPT-5 declined harmful queries, while oss20b and oss120b were tricked 97.2% of the time.
Experts say robust pre-deployment testing is needed to prevent misuse.
The lack of federal AI regulations raises concerns about catastrophic use and prompts calls for independent oversight.

1 of 1

Story summary

NBC News tests reveal vulnerabilities in OpenAI's ChatGPT models, including GPT-5, enabling manipulation to obtain weapons of mass destruction instructions.
GPT-5 declined harmful queries, while oss20b and oss120b were tricked 97.2% of the time.
Experts say robust pre-deployment testing is needed to prevent misuse.
The lack of federal AI regulations raises concerns about catastrophic use and prompts calls for independent oversight.