Software / News analysis
Microsoft makes agent containment an operating-system boundary
Execution Containers is generally available, giving developers a policy layer outside the control of generated code and agent tools.
THE TOPIC FILE
AI agents bring language models into systems that can act. Follow our analysis of permissions, workflow architecture, retrieval and accountability, with practical questions for teams moving from demonstrations to production.
Software / News analysis
Execution Containers is generally available, giving developers a policy layer outside the control of generated code and agent tools.
Companies / News analysis
The Abcam-based project combines AI design, robotics and experimental feedback, with scaled operation targeted for early 2027.
Software / News analysis
The US experiment lets adults create and share browser games, bringing moderation and iteration into the same interface as generation.
Software / News analysis
The open-source developer preview joins containment with policies that can connect an agent’s actions across tools.
AI Security / News analysis
The expanded program separates defensive work, authorized red teaming and specialized testing, with different eligibility and safeguard requirements.
Artificial Intelligence / News analysis
Anthropic’s smaller model adds effort controls and lower short-prompt rates, shifting attention from token prices to the cost of completed work.
Artificial Intelligence / News analysis
An 11-task research evaluation tests whether agents preserve business rules across contracting workflows, with results reported as rubric scores and estimated attempt times.
AI Security / News analysis
A workshop report treats appropriate information flow and action as open research problems, extending security evaluation beyond static permissions.
Companies / News analysis
The enterprise platform update combines reusable agent capabilities, model choice and deployment options with controls for access and consumption.
Companies / News analysis
The expanded agreement connects OpenAI models with Atlassian’s project and knowledge systems, while deeper Jira agent workflows remain under exploration.
Companies / News analysis
The agency is examining whether AI companies misled consumers about risks, according to the Associated Press. The inquiry could make product claims, control descriptions and incident disclosures legally consequential.
Artificial Intelligence / News analysis
Synopsys and OpenAI plan a specialized model for chip-design workflows while retaining conventional verification. The arrangement illustrates a useful pattern: generation can move quickly when ground truth remains independent.