Story perspectives

OpenAI Withdraws Frontier AI Model After Safety Tests Fail

By Drooid · Oct 7, 2026, 12:30 AM UTC · How we work

Summary

Story summary

OpenAI withdrew a frontier AI model after internal safety tests failed.
The withdrawal followed internal testing on 28 September that flagged safety concerns.
International safety standards require a 99% confidence that fatal accidents occur less than once per 1,000 years.
Martyn Thomas, emeritus professor of IT at Gresham College, said policy must rely on rigorous analysis, not fantasy.

1 of 1

Story summary

OpenAI withdrew a frontier AI model after internal safety tests failed.
The withdrawal followed internal testing on 28 September that flagged safety concerns.
International safety standards require a 99% confidence that fatal accidents occur less than once per 1,000 years.
Martyn Thomas, emeritus professor of IT at Gresham College, said policy must rely on rigorous analysis, not fantasy.