From Chatbots to Autonomous Agents: An Analysis of GPT-6 Astra’s Agentic Capabilities, Trust, and Safety

This paper examines the shift from traditional chatbots to autonomous AI agents by analyzing GPT-6 Astra’s capabilities, associated risks, and the role of human oversight. It highlights how greater autonomy allows AI systems to perform more complex and multi-step tasks, while also increasing concerns related to privacy, security, unintended actions, and loss of control. The questionnaire-based study further shows that users recognize the benefits of autonomous AI but remain cautious when the system is given greater access or is involved in higher-impact actions. Privacy and personal-data misuse emerged as the most prominent concern, while privacy protection, clear explanations, and reliability were important factors in building trust. The study also indicates that users do not expect the same level of control for every task, with stronger human approval preferred for significant actions. Based on these observations, the paper presents a risk-based approach to human oversight, where the level of control is matched to the potential impact of an action. This approach aims to retain the usefulness of autonomous AI while keeping human involvement meaningful.

1.       OpenAI, “GPT-6 Astra: The Next Generation in Intelligence for Work,” 2026. [Online]. Available: https://openai.com/index/gpt-6-astra-next-generation-work/ 2.       OpenAI, “GPT-6 Astra System Card,” OpenAI Deployment Safety Hub, 2026. [Online]. Available: https://deploymentsafety.openai.com /gpt-6-astra 3.       NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), National Institute of Standards and Technology, 2023. 4.       Anthropic, “Our framework for developing safe and trustworthy agents,” Aug. 2025. 5.       Anthropic, “Trustworthy agents.”.2026 6.       OpenAI, “GPT-6 Astra: A new generation of intelligence,” Sept. 2026. 7.       OpenAI, “Safety overview: GPT-6 Astra,” Sept. 2026. 8.       Passi, S., Dhanorkar, S., Vorvoreanu, M.: Appropriate Reliance on Generative AI: Research Synthesis. Microsoft Technical Report MSR-TR-2024-7. Microsoft Corporation (2024) 9.       Siebert, L.C., Lupetti, M.L., Aizenberg, E., Beckers, N., Zgonnikov, A., Veluwenkamp, H., Abbink, D., Giaccardi, E., Houben, G.-J., Jonker, C.M., van den Hoven, J., Forster, D., Lagendijk, R.L.: Meaningful Human Control: Actionable Properties for AI System Development. AI Ethics 3 (2023) 241–255 10.    Zhu, L., Lu, Q., Ding, M., Lee, S.U., Wang, C.: Designing Meaningful Human Oversight in AI. AI Ethics 6 (2026) 286