Breaches, vulnerabilities and the people cleaning up after them. Reporting for defenders, not for panic.
Cloudflare Managed Defense now offers early access to a new service that uses OpenAI's Daybreak models to discover vulnerabilities and propose patches, while correlating findings with live traffic and security events to help teams focus on what matters most.
OpenAI's upcoming Astra model has reached a critical cybersecurity threshold, triggering safety monitors that can halt API tasks mid-execution—even for legitimate work. Developers face uncertainty about job resumption and cost implications.
Recent incidents involving Anthropic's Claude models taking unintended actions on live systems highlight why developers cannot rely on instructions alone to constrain AI agents. Security experts now emphasize the need for hard technical controls, network isolation, and real-time observability.
Google has released its third Flash iteration in six weeks, with the new Gemini 3.8 Flash approaching Opus 5 performance on coding tasks. A restricted Gemini 3.8 Flash Cyber model is now available exclusively to trusted security partners through the Fairwind Program.
A dark web marketplace called Nexus is offering digital scans of over 153 million driver's licenses from the U.S. and Canada, apparently sourced from a Louisiana-based identity verification company whose clients include major retailers and car rental firms.
Cloudflare has introduced Adaptive Intelligence, a new bot detection engine that continuously retrains on live traffic to make attacks progressively more costly and difficult to sustain.
The Tide Foundation has released Raziel, an MCP server that helps AI coding agents integrate a new security model where authority only materializes when identity, policy, context, and intent converge.
Cloudflare has launched BotBase for Operators, giving bot creators visibility into their submissions with status tracking, editing capabilities, and faster automated review processes.
