Balyasny Asset Management evaluates and governs Claude Fable 5
Balyasny Asset Management (BAM), which oversees about $38 billion in assets with a staff of roughly 2,000, detailed how it tests and controls Anthropic’s Claude Fable 5. Chief AI Officer Charlie Flanagan explained that the firm runs thousands of real‑world financial tasks—covering equities, macro and commodities—to compare new models against its existing production model. In those tests Claude…
Key points
- Claude Fable 5 scored 89.4% versus 86.1% on BAM’s internal financial tasks.
- Agent workflow reduced merger‑arbitrage analysis from 3‑5 days to under one day (30‑minute run).
- BAMAgent runs thousands of autonomous agents 24/7, with over 300 agents handling continuous analysis.
The firm built its own platform, BAMAgent, to run autonomous agents that can perform multi‑step workflows for hours or days. A merger‑arbitrage analysis that previously required three to five days now completes in under one day, with the agent running about 30 minutes before human review. Other examples include a tax‑loss‑harvesting run that scanned 90,000 database tables and a central‑bank analysis cut from roughly two days to about 30 minutes. Safety is handled through data boundaries, least‑privilege access, tool permissions, logging, and mandatory human review, ensuring the model’s capabilities do not grant broader authority.
The story so far
4 episodes →- Balyasny Asset Management evaluates and governs Claude Fable 5this story
Working at the frontier: How Balyasny Asset Management evaluates and governs Claude Fable 5
Anthropic Engineering · 17 September 2026
Loading the full article…
This text was published by Anthropic Engineering. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Enterprise & Industry
All →- Anthropic replaces Git with Scion after AI agents drive merges from 2,500 to 30,000 per day · 11 src
- Sullivan & Cromwell builds legal tool Agreement Analyzer with OpenAI · 1 src
- Opsin raises $21M to track enterprise AI agents and flag risky behavior · 1 src
- Microsoft adds deepfake detection to Teams in November 2026 · 1 src
- Access Construction launches agentic AI layer Evo for estimating, project management and ERP · 1 src
Comments
via GitHub Discussions