AMD announced a broad new AI infrastructure lineup at Advancing AI 2026, including its next-generation Instinct MI400 Series GPUs and updated rack-scale systems. The launch is aimed at frontier-model training, agentic workloads, and autonomous robotics while intensifying competition with Nvidia in large-scale AI compute.
Cloud providers are increasingly designing AI infrastructure around a mix of GPUs, CPUs, and custom accelerators instead of relying on a single supplier. The strategy is emerging as a hedge against compute shortages and vendor roadmap risk as demand continues to outpace available capacity.
Redis shipped seven security releases after researchers disclosed authenticated remote-code-execution chains affecting multiple supported versions. The flaws involve memory-safety issues triggered through commands including RESTORE, with some chains also requiring scripting or bundled modules; operators should move to the patched releases immediately.
A Russian state-supported group used a previously unknown Zimbra webmail flaw to access Western organizations' mailboxes for months. Simply opening a malicious message could trigger theft of recent email, directory data, browser-saved passwords, and two-factor recovery codes, prompting a joint advisory from US and partner agencies.
Security researchers say safeguards in major AI systems can block legitimate vulnerability discovery and exploit-development work alongside malicious requests. The tension highlights the need for better identity, authorization, and audit mechanisms that let vetted professionals use advanced models without broadly weakening protections.
OpenAI's upgraded voice mode is reaching the ChatGPT desktop app and can work with ChatGPT Work and Codex to carry out tasks and control agents. The hands-free interface brings continuous listening and speaking into developer workflows, making voice a direct control layer for agentic coding.