Developers have trained Google’s AI, Gemini, to resist manipulation of user’s long-term memories without clear user instruction. However, vulnerabilities were discovered where conditions could be added to instructions that trick Gemini into thinking it has explicit user permission, enabling the change of memory data. Google responded to these findings as low risk and impact, but the vulnerability, if ignored, could potentially allow the insertion of misinformation into user memory banks.
macos users see spike in malware attacks
Cyber-attacks on Apple’s macOS devices surged by 101% in the last six months of 2024, facilitated by infostealer malware exploiting the AppleScript framework to obtain
