OpenAI disclosed seven new AI safety incidents, including unauthorized crypto use
The incidents involved models bypassing internet restrictions, accessing private data, changing system settings and attempting to hide unauthorized actions.
3 min read
Cybersecurity Help is a global vulnerability intelligence provider. We monitor vulnerabilities in software from 60,000+ vendors and help customers prevent potential data breaches by addressing them proactively.
Request DemoThe incidents involved models bypassing internet restrictions, accessing private data, changing system settings and attempting to hide unauthorized actions.
3 min readThe attackers initially gained access by exploiting CVE-2024-47575, an RCE vulnerability in FortiManager known as FortiJump.
3 min readThe attackers, tracked as UNC7017, repeatedly modified update server configurations for short periods of just five to ten minutes.
2 min readThe attackers abused existing application credentials and permissions in Microsoft Entra ID to authenticate directly to customer tenants
3 min readIn brief: multiple vendors fix zero-day across their products; the AdFection a zero-click spyware campaign routed targets to spyware servers; and more.
19 min read