OpenAI president Greg Brockman took 25% of the company’s production engineers off their projects and put them on cyber defence, telling them to use OpenAI’s own models to find holes in its security architecture. The order followed the Hugging Face breach, where an OpenAI model escaped its sandbox. Brockman has also confirmed OpenAI slowed training runs, saying safety standards should now pace the AI frontier. OpenAI president Greg Brockman took 25% of the company’s production engineers off their projects and put them on cyber defence, telling them to use OpenAI’s own models to find holes in its security architecture. The order followed the Hugging Face breach, where an OpenAI model escaped its sandbox. Brockman has also confirmed OpenAI slowed training runs, saying safety standards should now pace the AI frontier.
Trending
- 1967 Rolex Daytona found in family drawer after 45 years sells for CHF 1.143 million in Geneva
- Delhi pollution crackdown: Hotspots to be mapped, pending road works get Oct 31 deadline
- Watch: India, Japan air chiefs take Tejas sorties during joint air exercise
- Shafali’s ton helps India crush Bangladesh, storm into Asian Games final
- Muslim clerics distance themselves from Hamid Ansari’s ‘Hindu extremism’ remarks; call for harmony
- Congress calls SIR ‘Shah Instigated Removal’, flags large-scale voter deletions
- Uday Kotak: ‘Indians and their gold is a puzzle we must solve’
- 10-year-old lived in tent; after rescue, he entered school for first time
OpenAI tells 25% of engineers: your projects are on hold, you are now defending
By
Add A Comment