A framework for attributing responsibility in conversational AI agent chains: agent chains are modelled as directed acyclic graphs with provenance metadata and causal responsibility semantics, and harm is traced backwards to graded responsibility scores.
What a systematic evidence map of 626 peer-reviewed studies shows about AI in geothermal energy systems: how far methods have been validated beyond their development data, the E0–E4 evidence-readiness scale, and why transferability and deployment do not advance together.
How user consent breaks down when AI systems act on people’s behalf, interact with other agents and delegate to robots: the consent vacuum, consent chain degradation, the CoRVE runtime architecture and six consent vulnerabilities in agent ecosystems.
A deep reinforcement learning method for partially observable problems in which a guiding actor with privileged full-state access distils its policy into a control actor that acts on observation histories; extended in 2026 by Belief-Aware GSAC.