Microsoft Deploys 'Project Perception' to Corral Rogue AI Agents
Microsoft has expanded its Project Perception security suite to treat AI agents as potential internal threats, introducing dynamic permissions and real-time risk scoring for autonomous software. The system assigns every agent a continuous 'trust score' and automatically revokes access to critical systems if the agent's behavior deviates from its assigned task, requiring human intervention for high-stakes actions.
A cybersecurity report revealed that a single hacker used the open-source Cairn platform to autonomously probe and breach over 100 corporate websites with minimal human input.
New details on Anthropic's Public Benefit Fund show it will scale with the company's value, adding 2% more equity for every 50% jump in valuation ahead of its November IPO.
Notion rolled out a sixth-generation agent harness that can run single agent tasks for more than 24 hours (up from five minutes) and added a smarter model router called Auto that delegates work to the best-cost model.
Did you know?
A team of four Gemini 4 Argon agents successfully defeated the Minecraft Ender Dragon in 27 hours using only standard in-game tools, despite dying 605 times.
Quiz
Which major financial institution's CEO recently labeled outside AI agents 'too risky' and implemented a total block on their account access?