Microsoft aims for AI to identify hackers before they launch any attacks.

      Red team, blue team, green team: Microsoft has just equipped its AI with a comprehensive security team.

      Microsoft has recognized that with AI now capable of creating exploits and executing attacks at unprecedented speeds, traditional security tools designed for a slower, human-centric approach are becoming outdated. The company believes it has a solution, which it terms Project Perception.

      So, what is Project Perception?

      Essentially, Project Perception is an AI-driven security system designed to think and respond as rapidly as the threats it aims to combat. Instead of inundating security teams with an overwhelming number of alerts, it employs AI to monitor, analyze, and act upon the entire digital landscape of an organization continuously.

      The system comprises three types of AI agents working in collaboration. Red team agents identify vulnerabilities before they are exploited by attackers. Blue team agents evaluate these vulnerabilities to determine which ones are significant. Finally, green team agents address and rectify the identified issues. This collaboration creates a feedback loop that evolves and enhances over time, while ensuring that a human remains responsible for the final decisions.

      Why does Microsoft believe this method will succeed?

      Microsoft attributes its advantage to its visibility. The company can observe across various identities, devices, applications, data, and cloud infrastructures and, crucially, can take action based on these observations. This is complemented by a multi-model architecture that assigns the appropriate AI model for each specific task instead of depending on a single model for all functions.

      At its foundation lies what Microsoft refers to as its new cyber stack, a framework that converts raw signals into contextual information, which is then provided to models and agents capable of executing actions. It emphasizes practical responses over mere eye-catching alerts.

      An early instance of this is MAI-Cyber-1-Flash, a specialized model from Microsoft that is now operational within MDASH, its vulnerability management tool. Microsoft asserts that this model scores 96% on the CyberGym benchmark, outperforming Mythos by 12 points while reducing costs significantly in comparison to the existing system.

      Project Perception will enter public preview on August 3, and Microsoft asserts that it has designed the entire system with its Responsible AI principles from the outset. Cybersecurity has traditionally been a game of cat and mouse. With both offensive and defensive sides now leveraging AI, the key question is who can utilize it more effectively and swiftly.

      Rachit is an experienced tech journalist with over a decade of expertise in the consumer technology sector.

      AI can transform a cloud of scuba bubbles into an early warning system for divers.

      Researchers from the University of Minnesota have trained an AI underwater robot to gauge a diver’s breathing rate by observing the bubbles released during exhalation. This contactless system, detailed in a peer-reviewed article in The International Journal of Robotics Research, operates using standard camera footage, allowing divers to avoid adding another sensor to their already accessory-heavy equipment.

      Nike’s latest slide feels like a spa day for your feet.

      Nike has collaborated with Hyperice to create a slide that literally heats and vibrates your feet back to life. The new Nike Air Zoom Hyperslide feels more like a wellness device than traditional footwear. This slide is designed not just for relaxation by the pool but also to aid recovery after workouts, travel, or a long day of standing.

      What exactly is the Nike Hyperslide?

      If you shared a Claude chat link, Google may have encountered it too.

      Publicly shared chat links from Claude have been indexed by Google, potentially revealing sensitive information exchanged with the AI.

      If you've ever distributed a Claude conversation through a public link, it might be prudent to review your privacy settings. Recently, users found out that publicly shared Claude chats and Artifacts were showing up in Google Search and other search engines via queries like site:claude.ai/share. The indexed pages reportedly contained everything from resumes and business documents to API keys, health information, and other sensitive discussions that users likely did not anticipate would become searchable.

      How did this occur?