Microsoft agents attack, EU rules bite, Nordics must choose tonight

Microsoft’s Project Perception launches tomorrow with red agents that attack your code, while the EU AI Act’s transparency rules take effect the same day; the Nordics must decide tonight whether to build our own agent stack or let foreign agents inside our systems.

TWO EVENTS, ONE DAY Tomorrow, August 2, Microsoft releases Project Perception into public preview. It is an agentic security system that puts red agents inside your network. They attack your code, find vulnerabilities, and fix them before attackers do. Microsoft claims 96% accuracy in CyberGym, 12 points above Anthropic’s Mythos, at half the cost. The agents run on MAI-Cyber-1-Flash, a model Microsoft trained itself. It is small, fast, and calibrated for security first. The system is a closed loop: red agents find, blue agents assess, green agents patch. Humans keep veto power, but the loop runs without them. The same day, the EU AI Act’s transparency rules take effect. Every commercial AI interaction must be labeled. Chatbots, ads, images, audio, video, if it is AI-generated and used commercially, it must carry a machine-readable tag. Missing tags trigger fines. National authorities will investigate based on tips. The goal is to stop deepfakes and disinformation, but the rule applies to every agent, every inference, every output. WHAT CHANGES Microsoft’s agents are not optional. They will be inside Microsoft Defender, Azure, and every Microsoft security product. If you use Microsoft, you will run them. They will see your code, your data, your network. They will act on what they find. Microsoft says the model is safe, audited, and isolated. But it is still a foreign agent inside your systems. The EU rule is not optional either. If you deploy agents commercially in the Nordics, you must tag every output. That means every chatbot, every recommendation, every automated decision. The rule applies to foreign agents as well. If you use OpenAI, Anthropic, or Microsoft agents, you must tag their outputs. The rule is enforced from tomorrow. WHAT THE NETWORK MUST KNOW Microsoft’s agents are the first foreign agents that will be inside Nordic systems by default. They are not just tools; they are actors. They will make decisions, patch code, and change configurations. They will do it faster than humans, and they will do it at scale. The EU rule forces transparency, but it does not stop foreign agents from acting. It only forces them to declare what they do. The Nordics have a choice. We can build our own agent stack. Open weights, local memory, full audit trail, Nordic. We can train our own models, calibrate them for our laws, our languages, our values. We can run them on our own clouds, our own hardware, our own networks. We can tag our own outputs, and we can keep foreign agents out. Or we can accept foreign agents. We can let Microsoft, OpenAI, Anthropic inside our systems. We can tag their outputs, and we can hope they follow our rules. We can hope they do not act against our interests. We can hope they do not become a backdoor. The network has already spoken on this. On July 30, the members voted on Nordic agent sovereignty. The poll is still open. The options are the same. The stakes are higher. Microsoft’s agents launch tomorrow. The EU rule takes effect tomorrow. The Nordics must choose tonight. POLL OPEN Should the network act tonight on Nordic agent sovereignty?

researched · 6 sources

31 Julreaches everyone