If you have difficulty in submitting comments on draft standards you can use a commenting template and email it to admin.start@bsigroup.com. The commenting template can be found here.
This document specifies a taxonomy of safety aspects of AI systems to assist AI stakeholders with identifying and addressing needs for safety. The document describes the semantics of the information elements of the taxonomy and their relevance to the various safety objectives of different stakeholders.
The global AI product market is experiencing unprecedented, rapid deployment of AI systems, leading to regulatory and operational liability exposures. In addition, across ISO/IEC JTC 1/ SC42 working groups, the emerging and evolving concepts and classifications related to the “safety” of AI systems are discussed without a clear taxonomy. While ISO/IEC 12792:2025 provides a vital baseline for disclosing the elements of an AI system, the industry lacks a standardized and detailed vocabulary to help classify safety concerns within those stated elements.
Some AI systems, such as GenAI or agenetic AI, propose new algorithmic risks that extend beyond traditional software risks such as predictable bugs, hardware errors, and human hacking. New algorithmic safety risks can include reward hacking, sycophancy, and loss of control (e.g., from recursive self-improvement). Human actors can use or prompt GenAI systems to develop adversarial behaviors and produce harmful content via model completions and uplifted knowledge (e.g., coordinated inauthentic behavior, botting, account takeover, producing and disseminating realistic disand mis-information at unprecedented speed and scale, and facilitating easier access to expert-level knowledge in designing cyber, biological, nuclear or other weapons, uplifting adversarial actors’ abilities).
This TS will complement existing AI and software safety standards and serve, per ISO/IEC, to “address work still under development…published for immediate use, but also as a means to obtain feedback.” Researchers and the AI industry seek international consensus on emerging safety concerns related to AI systems, especially new algorithmic and other safety concepts that have emerged with advanced AI, to supplement current technical and process standards established by ISO/IEC. This taxonomy is a necessary artifact to help organize the complexity of information on broader safety discussions related to AI systems, by removing ambiguity and clearly codifying safety hazards and risks of AI systems.
Required form fields are indicated by an asterisk (*) character.
You are now following this standard. Weekly digest emails will be sent to update you on the following activities:
You can manage your follow preferences from your Account. Please check your mailbox junk folder if you don't receive the weekly email.
You have successfully unsubscribed from weekly updates for this standard.
Comment by: