1 CRIT · 6 HIGH · 6 MED · 1 INFO · THREAT RED · 14 items · Generated in 246s
Authentication bypass on an inference server can enable code execution, privilege escalation, data theft, and tampering.
Authenticated tenants could read or delete other tenants' workflow backups, exposing secrets and damaging integrity.
Unauthenticated users can invoke MCP tools using the operator's API key, enabling license changes and deletions under the operator account.
An authenticated tenant may access or delete another scope's workflow backups, exposing sensitive workflow data in multi-tenant mode.
A memory release flaw can let an attacker cause denial of service in the inference server.
A crafted request can crash Triton and disrupt model serving, causing denial of service.
Attackers can exhaust Triton server resources and disrupt model serving, causing denial of service.
A file descriptor handling flaw can let an attacker crash inference service availability.
A stack buffer overflow in Triton can crash inference services and disrupt model availability.
Attackers can exhaust server resources and cause inference outages or degraded availability.
A single email can poison an agent's stored memory and silently bias future replies across sessions.
A browser flaw may let other extensions read sensitive Gmail and Calendar data from Claude for Chrome sessions.
A poisoned repo can trigger unintended code execution in the IDE, putting developer systems and source code at risk.
This is a product and strategy announcement about hybrid cloud visibility, not a specific AI security vulnerability or exploit.