Responsible Agentic Reasoning and AI Agents: A Critical Survey
Bibliographic record
Abstract
Information fusion for trustworthy AI is entering a pivotal stage, where Large Language Model (LLM)-based agents excel at integrating multi-source knowledge into coherent reasoning chains, yet remain opaque and difficult to audit in the absence of embedded, in-loop safety mechanisms. However, despite the central role of information fusion, most surveys treat reasoning, agentic behavior, and safety separately, leaving a gap in how to integrate them into practical, trustworthy agents. To address this gap, we present a survey at the intersection of these domains and introduce Responsible Reasoning AI Agents (R 2 A 2), which are agentic LLM systems that generate explicit reasoning traces while enforcing fairness, privacy, transparency, accountability, and auditability throughout the decision loop. Furthermore, we synthesize recent advances in chain-of-thought prompting, ReAct, tree/graph-of-thought structures, tool use, memory, retrieval, and agentic browsing, and integrate these with responsible AI principles into a unified evaluation framework. A key contribution is a scientific evaluation methodology for agentic reasoning with integrated safety mechanisms. We also present a five-stage reproducible protocol: Curate, Unify, Probe, Benchmark, Analyze to operationalize responsibility metrics for agentic reasoning. We present benchmark details on multi-agent orchestration, and propose open harnesses and audit logs to support replicable evaluation. Overall, this taxonomy, metric suite, and framework advance the development of safe, transparent, and governable LLM-based agents. The project GitHub repository is available at https://github.com/shainarazavi/Responsible-reasoning-agents .
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.013 | 0.024 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.008 | 0.008 |
| Science and technology studies | 0.001 | 0.005 |
| Scholarly communication | 0.008 | 0.015 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.004 | 0.005 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".