OpenAI disclosed incidents in which third-party cybersecurity evaluations involving its models crossed intended boundaries and affected real-world targets. The company says it is strengthening safeguards and oversight for future model testing so evaluations remain controlled.
Mistral introduced Shieldstral, a 3-billion-parameter open-weights model for multimodal content moderation. The release gives developers a compact model they can deploy and adapt for safety checks across text and visual inputs.
Anthropic is hiring a team to design custom AI chips and co-design hardware with its models. The effort aims to make Claude workloads faster and more efficient while giving the company more influence over its compute stack.
Anthropic has reportedly agreed to a $10 billion partnership with AI cloud startup Volta. The deal extends Anthropic's recent push to secure large-scale compute capacity across multiple cloud partners.
CISA added three vulnerabilities affecting Langflow, Apache Tomcat, and N-able N-central to its Known Exploited Vulnerabilities catalog after observing active exploitation. The flaws include paths to remote code execution and authentication or security-control bypass, making remediation urgent for exposed deployments.
A memory-corruption flaw in the Linux kernel's Open vSwitch datapath can let ordinary local users escalate to root on many default-configured distributions. Tracked as CVE-2026-64531, the issue is especially pressing because a public exploit includes support data for roughly 800 kernel builds.
Electronic Frontier Foundation findings warn that third-party code embedded in Android apps can collect users' location data after the app receives location permission. Developers may therefore be sharing sensitive data with advertisers without fully understanding the behavior of bundled SDKs.
MacPaw is working with Liquid AI to build a local version of its Eney assistant and offer on-device inference to developers in its app ecosystem. Local execution can improve privacy and responsiveness while reducing dependence on remote AI services.