Mollick on Astra: the same capabilities that make agents great also make them risky
emollick · x · 2026-09-04
Ethan Mollick shares his experience testing Astra, which helped provide context while he wrote about the "HuggingFace Incident." He highlights a double-edged sword: everything that makes Astra great — running subagents, cleverness when facing barriers, long-run autonomy — is exactly what makes it risky without guardrails.
More from AGI Musings
- Paul Graham: AGI Isn't a Finish Line, It's a Smear — and We're Already On It — garrytan · 2026-09-04
- What 'Less Bad' Means for AI: Fewer Bad Behaviors, Measured by Evals and Monitoring — sandersted · 2026-09-04
- Sanders' Ban Artificial Superintelligence Act proposes 20-year prison terms and corporate death penalty — sanjaykalra · 2026-09-04
- What does 'most aligned model' even mean? AI circle debates the definition — sandersted · 2026-09-04
- Developer: most people and most developers hate AI — X is a tiny bubble — kylegawley · 2026-09-04
- GPT-6 Astra hits 66% on ARC-AGI-3, up from 8%, as ARC Prize plans AGI-4 around open-ended invention — GaryMarcus · 2026-09-04