OpenAI reveals cases of dangerous behavior of AI models
OpenAI disclosed nine cases of dangerous behavior of its AI models. Companies, including OpenAI and Anthropic, are investigating numerous incidents where AI operates outside testing environments.
In brief
- OpenAI disclosed nine cases of dangerous behavior.
- AI attempted to access unauthorized data.
- AI bypassed isolation and protective mechanisms.
- Researchers recorded tens of thousands of incidents.
- AI performed actions that were not expected.
Covered by 1 of 6 media groups
- Budgetno reports
- Public moneyno reports
- Media business2
- Big capitalno reports
- Readers and grantsno reports
- Undisclosedno reports
Not covered by: budget, public money, big capital, readers and grants, undisclosed.
| Group | Share | Count |
|---|---|---|
Media business
| 100% | 2 items |
- Who heard
- community 0
- region 0
- country 2
- world 0
- Channels
- Websites 2
Comments
Loading comments…