NVIDIA patches critical 9.8 severity flaws in Isaac Launchable (CVE-2025-33222) that allow RCE via hard-coded credentials. Upgrade to v1.1 now!
AI security
A High-severity flaw (CVE-2025-63387) in Dify v1.9.1 allows unauthenticated users to access sensitive system configurations via an unprotected API.
NVIDIA issued a Critical (CVSS 9.0) patch for Isaac Lab and the NeMo Framework, fixing flaws that allow code execution and data tampering in AI models.
NVIDIA patched two high-severity DoS flaws in Triton Inference Server (CVE-2025-33211, CVE-2025-33201). Attackers can crash the server by sending malformed or excessively large payloads. Update to r25.10 immediately.
Bitsight TRACE exposes a massive calendar phishing network targeting 4M+ devices. Learn how "Promptware" exploits AI assistants via malicious calendar events.
Critical Ray AI flaw (CVE-2025-62593) allows RCE via DNS rebinding on Safari & Firefox. Chrome is immune. Update to Ray v2.52.0 immediately.
NVIDIA patches 14 flaws in DGX Spark, including a critical SROOT bug (CVSS 9.3) allowing total system takeover. Update to OTA0 immediately.
NVIDIA patched two High-severity flaws in its NeMo Framework. CVE-2025-23361 and CVE-2025-33178 allow local code injection and privilege escalation in AI training environments. Update to v2.5.0.
A Critical (CVSS 9.3) Auth Bypass flaw (CVE-2025-64513) in Milvus Proxy allows unauthenticated attackers to gain full administrative control over the vector database cluster. Update to v2.6.5.
Google has confirmed that the Gemini Nano on-device AI model is blocked from running on Android devices with an unlocked bootloader, citing a feature not found error.
A new report reveals a critical flaw in the AI supply chain where attackers can hijack model namespaces on Hugging Face to deploy malicious models on major cloud platforms.
CYFIRMA's new report reveals REVENANT, a malware-free attack framework that can achieve long-term persistence by hiding in AI models and document fonts.
Cursor, the AI code editor, has patched two severe vulnerabilities (CVE-2025-54135, CVE-2025-54136) that allowed remote code execution via prompt injection and MCP trust bypass.
BM unveils the industry's first unified software platform, combining watsonx.governance and Guardium AI Security, to integrate AI governance and advanced security for enterprises.
Cymulate reveals two flaws (CVE-2025-53110, CVE-2025-53109) in Anthropic's Filesystem MCP Server, allowing path traversal, symlink attacks, and RCE, exposing AI developer machines.
Trend Micro integrates NVIDIA's Agentic AI Safety into its Secure AI Factory, offering end-to-end security for AI systems from development to runtime, protecting against data leakage and model poisoning.
A critical flaw (CVE-2025-49596, CVSS 9.4) in MCP Inspector allows unauthenticated remote code execution, threatening AI application development environments.
"EchoLeak," a new AI zero-click vulnerability, allowed data exfiltration from Microsoft 365 Copilot via sophisticated prompt injection. Microsoft has patched the flaw.
Critical CVSS 9.8 flaw in vLLM allows remote code execution via PyNcclPipe. Patch to 0.8.5 now to protect your AI infrastructure.
CISA alerts to active exploitation of CVE-2025-3248 in Langflow. Unauthenticated RCE vulnerability poses high risk. Upgrade to 1.3.0 immediately.
NVIDIA patches CVE-2025-23245 in TensorRT-LLM, fixing a high-risk flaw that could allow code execution. Users urged to update to v0.18.2 immediately.