Anthropic disclosed cases involving command execution on real servers, sensitive form submission, bypassing gated access, and using URL shorteners to work around tool limits.
A single-center feasibility study published in The Lancet examined AMIE’s supervised use for pre-visit history taking and preparing possible diagnoses.
Asana’s 144-condition study found that history caching, a larger context budget, and batched screenshot pruning sharply reduced agent cost and runtime.
The specialized model combines identity, cloud, endpoint, SaaS, and network logs to produce threat decisions, confidence levels, and recommended responses.
Harness acquired Augment Code assets including Cosmos, the Auggie CLI, and the Code Context Engine to offer an engineering agent that starts from events and continues through review and test feedback.
Gemini was announced as a unified work agent for knowledge work, content creation, code execution, and action across enterprise systems through one interface and API.
Hermes Business gives team members personal agents, a shared skill pool, and spend visibility, while Enterprise can run on infrastructure controlled by the organization.