CISA flags active exploits in OpenSSH, GPUStack, and IBM flaws
CISA added four vulnerabilities to its KEV catalog due to active exploitation, including the SSHole bug in OpenSSH and critical flaws in GPUStack and IBM Aspera.
Daily coverage of AI, developer tools and infrastructure. Each story explains what happened and why it matters, with a link to the original source.
CISA added four vulnerabilities to its KEV catalog due to active exploitation, including the SSHole bug in OpenSSH and critical flaws in GPUStack and IBM Aspera.
Google released TPU Sync, an open-source system that cuts time to first token by up to 90% for large models on Cloud TPUs.
Cursor and Google now let developers speak with coding agents in real time, while Anthropic’s Claude Code remains limited to dictation.
Google has updated its privacy policy to confirm that personal content from Gmail and Drive connected to consumer Gemini may be reviewed by humans, reversing a 2023 promise of isolation.
New research shows VMs still run over half of production apps, with AI workloads increasing demand for virtualization-based security and isolation.
CISA has flagged a SonicOS vulnerability that allows attackers to bypass SAML multi-factor authentication, citing reports of active exploitation.
Meta engineers introduced a new accelerator plugin architecture in lldb-server that unifies CPU and GPU debugging sessions while keeping hardware-specific logic separate.
US officials allege Nvidia CEO Jensen Huang had a personal relationship with an executive believed to be a Chinese intelligence asset, raising national security concerns.
Amazon Bedrock Managed Knowledge Base now supports single-shot RAG and retrieval-only requests via the AgenticRetrieveStream API, allowing developers to balance latency and accuracy.
Amazon SageMaker JumpStart now hosts NVIDIA’s new open-weight vision-language model, featuring a 120B parameter architecture and optimized inference for document and video analysis.
Attackers used stolen CI credentials to publish malicious versions of bee-js and Swarm Desktop, installing backdoors and stealing node wallet keys.
Meta Superintelligence Lab released Muse Ember, a lightweight AI model optimized for cost-efficient agent execution and multimodal parsing at scale.
The CNCF Sandbox project Hyperlight now runs standard Linux binaries in hypervisor-isolated micro-VMs, using snapshots to achieve millisecond startup times for AI agents and plugins.
A new open-weight model, Nenya-2.5B, demonstrates competitive performance on small hardware using GDN-2 layers and flash-linear-attention kernels for faster inference.
Anthropic released Claude Fable 5.2 with a 20% price cut, faster token generation, and improved coding benchmarks for complex technical tasks.