Research question
What harms become more likely as systems become more agentic?
Key findings
- Characterises agency by underspecification, directness, goal-directedness and long-term planning - Anticipates systemic, collusive and deceptive harms
Harms from Increasingly Agentic Algorithmic Systems
A FAccT analysis anticipating agent harms before LLM agents were widespread.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.