Research

Anthropic Hosts 'The Briefing: AI for Science' in San Francisco

Anthropic held 'The Briefing: AI for Science' in San Francisco on June 30, 2026, bringing together company leadership, life-sciences executives, and leading research institutions to show how Claude is being used across scientific research, with a public livestream at 9:00am PST. The event underscores Anthropic's growing push into AI-for-science and wet-lab partnerships.

Anthropic's 'Cadences' Economic Index Report Links Claude Usage to Daily Rhythms

Anthropic published its Economic Index report 'Cadences,' showing how the rhythms of the external world shape Claude usage — work queries subside on weekends, news requests peak in the morning, sleep-advice requests peak around 5 a.m., and tax requests surge near filing deadlines. The report draws on chat and Cowork conversations sampled April 10–June 10, 2026, and links about 9,700 survey respondents to their usage data using privacy-preserving methods, finding most respondents expect significant AI progress over the next year.

Google DeepMind and A24 Form AI Filmmaking Research Partnership

Google DeepMind and independent studio A24 announced a partnership to explore how artificial intelligence can support filmmakers and creative professionals. The agreement brings a frontier AI laboratory into direct research collaboration with a major film studio.

CLI-Universe Releases 6,000 Verified Terminal-Agent Trajectories

Researchers introduced CLI-Universe, a synthesis and verification pipeline for executable terminal-agent tasks. Fine-tuning Qwen3-32B on its 6,000-trajectory dataset achieved 33.4% on Terminal-Bench 2.0, reported as a new open-data result for models at or below 32 billion parameters.

Researchers Propose Intent-Governed Authorization for AI Agents

A new paper proposes Intent-Governed Access Control, a server-side authorization layer that checks whether tool calls and payloads remain consistent with the user's expressed request. The design ensures that inferred intent can narrow existing permissions but cannot expand them.

VISTA Architect Uses Provenance Graphs for Tumor Board Preparation

VISTA Architect converts longitudinal electronic health records into a provenance-linked clinical knowledge graph for repeated downstream use. In a thoracic oncology demonstration across 1,180 patients, it reported 96.4% accuracy on 17,700 evaluations and prepared a 30-patient cohort in about 2.2 minutes.