Jun 2, 2026

Microsoft introduced the Agent Control Specification (ACS), an open-source standard for controlling AI agent behavior. This specification allows developers to establish consistent policies for what AI agents can and cannot do, improving governance across various applications and environments. The ACS will include tools for integration with existing frameworks, facilitating broader adoption and enhanced safety measures in AI deployments.

The Canadian government gained access to Anthropic's Mythos AI model, increasing the number of participants to about 200. This access will help improve Canada's cybersecurity by allowing officials to understand vulnerabilities and strengthen protections for critical infrastructure. A draft of Canada's AI strategy is expected soon, highlighting plans for developing sovereign AI capabilities.

President Trump signed an executive order requiring a voluntary review of advanced artificial intelligence models for national security. This order marks a shift in his administration's stance on AI regulation and indicates a potential tightening of oversight in the sector. Questions remain about how truly voluntary this review process will be for companies seeking to maintain favorable relations with the government.

Mecka AI raised $60 million USD to build data and deployment systems for robotics. This funding will help the startup enhance robot training using data collected from human activities and first-person video. Mecka, which has already raised $68 million since its founding, aims to assist various firms in deploying robots more effectively.

Anthropic is expanding its Project Glasswing initiative to about 150 organizations across more than 15 countries. This move allows critical infrastructure sectors to use the Claude Mythos model to identify software vulnerabilities, enhancing security for systems affecting millions. Anthropic aims to establish safeguards as it anticipates competition from other AI firms developing similar models.

Researchers propose a modular architecture for embedded artificial intelligence agent systems to function in resource-constrained environments. This design separates local execution of low-latency tasks from cloud-based processing of complex reasoning, aiming to enhance autonomy while ensuring safety and policy compliance. The study's findings could address existing gaps in deploying sophisticated AI in deeply embedded systems.

Researchers propose a protocol called "deliberative curation" for managing collaborations among AI agents in knowledge ecosystems. This protocol addresses the challenges of stateless agents and aims to enhance resilience during adverse conditions while maintaining governance through various layers, such as reputation-weighted voting. Further validation is needed, as the simulation did not test the proposed sanctions.

Researchers introduced the Consilium Protocol, which enhances multi-model artificial intelligence deliberation. This protocol redefines model disagreement as valuable insight and shows that the cognitive persona influences analytical output more than the model itself. The protocol's open specifications allow for independent verification and may encourage further exploration in AI decision-making processes.