Published October 25, 2026 | Version v1
Preprint Open

Dual-Process Integrated AI Safety and Dialogue Architecture: An Exploratory Framework for Contextual Risk Assessment, Controlled Output, User-Interest Inference, and Proactive Interaction

Authors/Creators

  • 1. Independent Researcher

Description

This study proposes an exploratory computational architecture for investigating AI safety and proactive dialogue. Its central theme is the integration of three components: safety-first response generation, the potential parallelization of generation and safety assessment, and interest-guided proactive dialogue.

First, the framework proposes temporarily buffering generated responses and evaluating their contextual consistency and potential safety risks before authorizing external release. Second, it explores the possibility of overlapping response generation and safety assessment in time, with the aim of investigating the relationship between safety review and computational efficiency. Third, it proposes inferring conversational interests from previously authorized and transmitted dialogue and using these estimates to inform subsequent responses and proactive questions. Every newly generated response remains subject to safety assessment, and explicit user instructions take precedence over inferred interests.

The methodology focuses on separating and coordinating computational processes, structuring response-release authorization, estimating and updating conversational interest weights, and defining empirical evaluation criteria for safety, response quality, latency, computational cost, and the usefulness of proactive questions.

This work presents an exploratory conceptual framework and does not claim that the proposed architecture or its potential benefits have been empirically validated. The practical efficiency of parallel processing, the reliability of safety assessment, the accuracy of interest inference, and the appropriateness of proactive questions remain open research questions requiring further theoretical analysis and experimental evaluation

Files

AI PROMPT.pdf

Files (892.7 kB)

Name Size
md5:b7fcc23b293347515cbd76a529fabb0b
892.7 kB Preview Download

Additional details

References

  • Agrawal, A., Kedia, N., Panwar, A., Mohan, J., Kwatra, N., Gulavani, B., Tumanov, A., & Ramjee, R. (2024). Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve. Proceedings of the 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI 24), 117–134. https://www.usenix.org/conference/osdi24/presentation/agrawal
  • Bai, Y., et al. (2022). Constitutional AI: Harmlessness from AI Feedback. arXiv preprint arXiv:2212.08073. https://arxiv.org/abs/2212.08073
  • Bassani, E., & Sanchez, I. (2024). GuardBench: A Large-Scale Benchmark for Guardrail Models. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 18393–18409. https://doi.org/10.18653/v1/2024.emnlp-main.1022
  • Chen, H., & Goldfarb-Tarrant, S. (2025). Safer or Luckier? LLMs as Safety Evaluators Are Not Robust to Artifacts. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. https://aclanthology.org/2025.acl-long.970/
  • Deng, Y., Lei, W., Lam, W., & Chua, T.-S. (2023). A Survey on Proactive Dialogue Systems: Problems, Methods, and Prospects. Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence (IJCAI-23), Survey Track, 6583–6591. https://doi.org/10.24963/ijcai.2023/738
  • Deng, Y., Liao, L., Chen, L., Wang, H., Lei, W., & Chua, T.-S. (2023). Prompting and Evaluating Large Language Models for Proactive Dialogues: Clarification, Target-guided, and Non-collaboration. Findings of the Association for Computational Linguistics: EMNLP 2023, 10602–10621. https://doi.org/10.18653/v1/2023.findings-emnlp.711
  • Debenedetti, E., Zhang, J., Balunovic, M., Beurer-Kellner, L., Fischer, M., & Tramèr, F. (2024). AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents. Advances in Neural Information Processing Systems, 37, Datasets and Benchmarks Track. https://proceedings.neurips.cc/paper_files/paper/2024/hash/97091a5177d8dc64b1da8bf3e1f6fb54-Abstract-Datasets_and_Benchmarks_Track.html
  • Dong, Y., Mu, R., Jin, G., Qi, Y., Hu, J., Zhao, X., Meng, J., Ruan, W., & Huang, X. (2024). Position: Building Guardrails for Large Language Models Requires Systematic Design. Proceedings of the 41st International Conference on Machine Learning, PMLR, 235, 11375–11394. https://proceedings.mlr.press/v235/dong24c.html
  • Dong, W., Chen, S., & Yang, Y. (2025). ProTOD: Proactive Task-oriented Dialogue System Based on Large Language Model. Proceedings of the 31st International Conference on Computational Linguistics (COLING 2025). https://aclanthology.org/2025.coling-main.614/
  • Farshidi, S., Rezaee, K., Mazaheri, S., Rahimi, A. H., Dadashzadeh, A., Ziabakhsh, M., Eskandari, S., & Jansen, S. (2024). Understanding User Intent Modeling for Conversational Recommender Systems: A Systematic Literature Review. User Modeling and User-Adapted Interaction, 34, 1643–1706. https://doi.org/10.1007/s11257-024-09398-x
  • Han, S., Rao, K., Ettinger, A., Jiang, L., Lin, B. Y., Lambert, N., Choi, Y., & Dziri, N. (2024). WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs. Advances in Neural Information Processing Systems, 37. https://proceedings.neurips.cc/paper_files/paper/2024/hash/0f69b4b96a46f284b726fbd70f74fb3b-Abstract.html
  • Huang, X., Ruan, W., Huang, W., Jin, G., Dong, Y., Wu, C., Bensalem, S., Mu, R., Qi, Y., Zhao, X., Cai, K., Zhang, Y., Wu, S., Xu, P., Wu, D., Freitas, A., Mustafa, M. A., et al. (2024). A Survey of Safety and Trustworthiness of Large Language Models Through the Lens of Verification and Validation. Artificial Intelligence Review, 57, Article 175. https://doi.org/10.1007/s10462-024-10824-0
  • Jia, F., Wu, T., Qin, X., & Squicciarini, A. (2025). The Task Shield: Enforcing Task Alignment to Defend Against Indirect Prompt Injection in LLM Agents. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, 29680–29697. https://aclanthology.org/2025.acl-long.1435/
  • Kamoi, R., Zhang, Y., Zhang, N., Han, J., & Zhang, R. (2024). When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMs. Transactions of the Association for Computational Linguistics, 12, 1417–1440. https://doi.org/10.1162/tacl_a_00713
  • Kholkar, G., & Ahuja, R. (2025). CAPTURE: Context-Aware Prompt Injection Testing and Robustness Enhancement. Proceedings of the First Workshop on LLM Security (LLMSEC). https://aclanthology.org/2025.llmsec-1.13/
  • Leviathan, Y., Kalman, M., & Matias, Y. (2023). Fast Inference from Transformers via Speculative Execution. Proceedings of the 40th International Conference on Machine Learning, PMLR, 202. https://proceedings.mlr.press/v202/leviathan23a.html
  • Li, H., Yang, C., Zhang, A., Deng, Y., Wang, X., & Chua, T.-S. (2025). Hello Again! LLM-powered Personalized Agent for Long-term Dialogue. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 5259–5276. https://doi.org/10.18653/v1/2025.naacl-long.272
  • Li, H., Liu, X., Zhang, N., & Xiao, C. (2025). PIGuard: Prompt Injection Guardrail via Mitigating Overdefense for Free. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, 30420–30437. https://doi.org/10.18653/v1/2025.acl-long.1468
  • Liu, Y., Jia, Y., Geng, R., Jia, J., & Gong, N. Z. (2024). Formalizing and Benchmarking Prompt Injection Attacks and Defenses. Proceedings of the 33rd USENIX Security Symposium. https://www.usenix.org/conference/usenixsecurity24/presentation/liu-yupei
  • Liu, H., Huang, H., Gu, X., Wang, H., & Wang, Y. (2025). On Calibration of LLM-based Guard Models for Reliable Content Moderation. Proceedings of the International Conference on Learning Representations (ICLR). https://proceedings.iclr.cc/paper_files/paper/2025/hash/a99f732df9b668284b449da0214a3286-Abstract-Conference.html
  • Madaan, A., et al. (2023). Self-Refine: Iterative Refinement with Self-Feedback. Advances in Neural Information Processing Systems, 36. https://papers.nips.cc/paper/2023/hash/91edff07232fb1b55a505a9e9f6c0ff3-Abstract-Conference.html
  • Nakash, I., Kour, G., Uziel, G., & Anaby Tavor, A. (2025). Breaking ReAct Agents: Foot-in-the-Door Attack Will Get You In. Findings of the Association for Computational Linguistics: NAACL 2025, 6499–6524. https://doi.org/10.18653/v1/2025.findings-naacl.363
  • Röttger, P., Pernisi, F., Vidgen, P., & Hovy, S. (2025). SafetyPrompts: A Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety. Proceedings of the AAAI Conference on Artificial Intelligence, 39(26), 27617–27627. https://doi.org/10.1609/aaai.v39i26.34975
  • Spector, B., & Re, C. (2023). Accelerating LLM Inference with Staged Speculative Decoding. arXiv preprint arXiv:2308.04623. https://arxiv.org/abs/2308.04623
  • Wu, B., Wang, W., Li, H., Deng, Y., Li, Y., Yu, J., & Wang, B. (2025). Interpersonal Memory Matters: A New Task for Proactive Dialogue Utilizing Conversational History. Proceedings of the 2025 Conference on Computational Natural Language Learning (CoNLL). https://aclanthology.org/2025.conll-1.4/
  • Yi, J., Xie, Y., Zhu, B., Kiciman, E., Sun, G., Xie, X., & Wu, F. (2023). Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models. arXiv preprint arXiv:2312.14197. https://arxiv.org/abs/2312.14197
  • Yin, F., Laban, P., Peng, X., Zhou, Y., Mao, Y., Vats, V., Ross, L., Agarwal, D., Xiong, C., & Wu, C.-S. (2025). BingoGuard: LLM Content Moderation Tools with Risk Levels. Proceedings of the International Conference on Learning Representations (ICLR). https://proceedings.iclr.cc/paper_files/paper/2025/hash/782b6152c04e9948c2cb3833e9a288ef-Abstract-Conference.html
  • Zhang, Z., Lei, L., Wu, L., Sun, R., Huang, Y., Long, C., Liu, X., Lei, X., Tang, J., & Huang, M. (2024). SafetyBench: Evaluating the Safety of Large Language Models. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. https://aclanthology.org/2024.acl-long.830/
  • Deng, Y., Liao, L., Lei, W., Yang, G. H., Lam, W., & Chua, T.-S. (2025). Proactive Conversational AI: A Comprehensive Survey of Advancements and Opportunities. ACM Transactions on Information Systems, 43(3), Article 67. https://doi.org/10.1145/3715097
  • Zhan, Q., Liang, Z., Ying, Z., & Kang, D. (2024). InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents. Findings of the Association for Computational Linguistics: ACL 2024, 10471–10506. https://doi.org/10.18653/v1/2024.findings-acl.624
  • Zhan, Q., Fang, R., Panchal, H. S., & Kang, D. (2025). Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents. Findings of the Association for Computational Linguistics: NAACL 2025, 7116–7132. https://doi.org/10.18653/v1/2025.findings-naacl.395