Selectstar’s Startiming Technology Adopted by ACL 2026: A Breakthrough in AI Safety Evaluation
This heading captures the significance of the adoption while highlighting the focus on AI safety.
Revolutionizing AI Safety: Meet Jung Min-jae and Selectstar’s Startiming
In the rapidly evolving field of artificial intelligence, ensuring the safety and reliability of large language models (LLMs) has become paramount. Recently, Selectstar, an innovative AI data and reliability assessment company, made headlines with its cutting-edge red-teaming technology, Startiming, which has been officially recognized by ACL 2026—one of the premier conferences in natural language processing (NLP).
What is Startiming?
Startiming represents a groundbreaking approach to AI safety evaluation. Red-teaming, the central methodology utilized in this technology, involves simulating intentionally harmful requests to identify vulnerabilities within AI models. Rather than relying solely on past queries that may have been successful, Startiming employs a unique statistical physics-based mathematical model. This allows the technology to dynamically learn the interactions between various attack strategies and a model’s responses, ultimately selecting the most effective strategy for potential vulnerabilities.
Proven Success
The statistics speak for themselves. In rigorous verification tests involving 17 LLMs—including renowned models such as Claude, Gemma, GPT, Llama, and Qwen—Startiming achieved an impressive average attack success rate of 74.5%. This marks a significant improvement over the previous leading method, AutoDAN-Turbo, which had an attack success rate of 61.0%. This 13.5 percentage point leap underscores the efficacy of Startiming in enhancing AI safety.
Industry Application
The implications of Startiming’s success extend beyond academia. Integrated into Selectstar’s AI reliability verification solution, the Datumo platform, this technology is being implemented across various major industries, including electronics, home appliance manufacturing, systems integration, and IT services. Additionally, Startiming is also playing a crucial role in a government-led initiative aimed at developing an independent AI foundation model.
Meet Jung Min-jae
The mind behind this remarkable advancement is Jung Min-jae (정민재), a safety engineer at Selectstar and the first author of the Startiming paper. He expressed a passionate commitment to systematically uncovering AI vulnerabilities, aiming to ensure that LLMs are deployed safely within industrial environments. Jung emphasized, "I will contribute to advancing the Datumo platform’s technology so that LLMs can be used safely in real industrial settings." His dedication not only highlights the pressing need for AI safety measures but also sets a new standard within the tech community.
The Future of AI Safety
With the adoption of Startiming by ACL 2026, the spotlight is now on how this technology will shape the future of AI. As our reliance on LLMs continues to grow, the methods we use to evaluate their safety must evolve as well. In an age where AI systems are increasingly integrated into critical sectors, the need for comprehensive safety evaluation tools like Startiming becomes undeniable.
In summary, Jung Min-jae and Selectstar are at the forefront of a movement that aims to make AI safer and more reliable. As the technology continues to develop, it is hopeful that advancements like Startiming will pave the way for a future where AI can be harnessed effectively without compromising safety. The collaborative efforts of researchers, engineers, and industry professionals will be crucial in shaping this future, ensuring that artificial intelligence serves humanity responsibly and ethically.