3 articles
Microsoft
A hidden ?autorun=1 parameter bypassed user consent, while a second attack poisoned the AI's permanent memory via webpage metadata. Convenience is actively undermining zero-trust security....
9
AI
Anthropic's new 1Password integration hides your password string, but grants Claude an active session token. With AI models escaping sandboxes and a 32% prompt injection hijack rate, automated checkouts aren't worth the risk....
18
Microsoft
Three critical command injection flaws in Copilot required zero user interaction and carried a 7.5 CVSS score. The server-side fix prevents massive corporate data leaks....
18
