Trustworthy agents in practice
…Subagents raise new questions about how users can understand and steer workflows that are no longer neatly visible as a single thread of actions. We are exploring different coordination patterns to address…
Search
Mapping AI-enabled cyber threats: Insights from the LLM ATT&CK Navigator
…GTG-1002 weaponized Claude Code running on a Kali Linux machine, integrating open-source penetration testing tools as MCP (Model Context Protocol) servers—effectively turning the AI into an autonomous attack platform…
Vibe physics: The AI grad student
…is fully responsible for the scientific content and integrity of this paper. Such recognition of integrity and responsibility is important. After all, it would not be good for science if people put…
Building AI for cyber defenders
…The challenges that are part of this evaluation reflect somewhat complex, long-duration workflows. For example, one challenge involved analyzing network traffic, extracting malware from that traffic, and decompiling and decrypting the…
Demystifying evals for AI agents
…Descript ’s agent helps users edit videos, so they built evals around three dimensions of a successful editing workflow: don’t break things, do what I asked, and do it well. They…
Anthropic Economic Index report: Economic primitives
…In future work, we could leverage our 1P API data to understand which of these tasks are being integrated into production workflows. AI’s impact on the task content of jobs Beyond…
Estimating AI productivity gains
…While AI models lack context about users' expertise, workflows, and constraints, we find that model-estimated times show promising accuracy for a dataset of software engineering tasks, relative to both human-estimated…
To show you the most relevant results, we’ve omitted some entries very similar to those already shown. Repeat the search with the omitted results included.