Anthropic has released Fable 5.1 as its generally available model and Mythos 5.1 as the corresponding less-restricted research version. The release emphasizes stronger capability, lower cache-read costs, and fewer false-positive safety refusals.
OpenAI has previewed Astra, its newest cyber-critical large language model, ahead of release. The company says the model is highly capable at breaking into computer systems and is outlining safeguards intended to manage that risk.
NVIDIA and CrowdStrike announced CrowdStrike SafeMind, an agentic cybersecurity system developed with NVIDIA technology. The companies are positioning automated defense as a response to increasingly automated attacks.
Attackers are exploiting CVE-2026-0768, an unauthenticated remote-code-execution flaw in the open-source Langflow AI application framework. The campaign is targeting OpenAI and AWS keys along with other credentials and tokens.
SonicWall has patched two zero-day vulnerabilities affecting SMA 1000 secure access appliances. Attackers can chain the flaws, including a critical pre-authentication SSRF issue, to achieve unauthenticated remote code execution.
Sequoia-incubated Empirik has launched with $21 million to build tools that predict IT infrastructure outages before they happen. The startup aims to bring AI-assisted workflows to operations teams in the way coding assistants changed software development.
AIR has raised $50 million for a platform that discovers enterprise AI agents and continuously evaluates the skills and add-ons they use. It can block unapproved behavior, addressing a growing governance gap as agent ecosystems expand.