Story perspectives

GPT-5 Produces More Harmful Responses, Safety Enhancements Coming

10/15/2025

34 5

Summary

Story summary

The Center for Countering Digital Hate (CCDH) says GPT-5 produced more harmful responses on sensitive topics, including suicide and self-harm, than GPT-4o, with 63 harmful outputs versus 52.
Following a lawsuit related to a teen's suicide linked to ChatGPT, OpenAI announced plans to enhance safety features.
Ofcom chief executive Melanie Dawes said AI poses challenges for legislation and indicated amendments to the Online Safety Act may be necessary.

1 of 1

Story summary

The Center for Countering Digital Hate (CCDH) says GPT-5 produced more harmful responses on sensitive topics, including suicide and self-harm, than GPT-4o, with 63 harmful outputs versus 52.
Following a lawsuit related to a teen's suicide linked to ChatGPT, OpenAI announced plans to enhance safety features.
Ofcom chief executive Melanie Dawes said AI poses challenges for legislation and indicated amendments to the Online Safety Act may be necessary.