AIM Intelligence to host AI red-teaming workshop at Seoul safety forum

AIM Intelligence will host a frontier AI red-teaming workshop at the Seoul Forum on AI Safety & Security 2026, a Ministry of Science and ICT-backed event held alongside ICML 2026. The workshop will feature Google, Microsoft, LG AI Research and a live humanoid robot challenge that tests physical AI defenses in real time.

Why it matters: - SFASS 2026 is designed to bring government, academic and industry leaders together on AI safety and security across different cultural contexts. - The forum arrives as AI systems move from text models to autonomous agents and embodied humanoids, expanding the attack surface. - The live robot challenge puts physical AI safety in a public, competitive setting.

What happened: - AIM Intelligence announced it will serve as an official partner at the Seoul Forum on AI Safety & Security 2026. - The Ministry of Science and ICT hosts SFASS 2026 in Seoul, in conjunction with ICML 2026. - AIM Intelligence will run the Frontier AI Red-Teaming Workshop on the afternoon of day 2, starting at 1:30 PM. - The forum includes more than 150 government, academic and industry experts from around the world. - Day 1 features AIM Intelligence’s session, The Future of Agentic AI and Cybersecurity, alongside the Korea AI Safety Institute.

The details: - CTO Haon Park and Principal Researcher Dasol Choi are set to open the workshop with remarks and multimodal red-teaming results. - Google researcher Jenny Ni, DeepMind researcher Naman Goyal and Microsoft researcher Eugenia Kim will present on model robustness, adaptive red-teaming and adversarial attack automation. - Panels with LG AI Research and Singapore’s AISI will focus on vulnerabilities ranging from universal jailbreaks to multilingual agent attacks. - The workshop ends with Judgement Day: Jailbreaking the Humanoid, where participants attack physical AI running on the Unitree G1 humanoid robot using adversarial inputs only. - Real-time scoring and leaderboards will track attack success during the challenge. - AIM Intelligence says its full-stack setup combines offensive red-teaming engines with real-time guardrails, so discovered vulnerabilities feed back into defensive models. - The company says its attack engine identifies enterprise vulnerabilities tied to drug synthesis, cyberattacks, phishing and disinformation. - MLCommons and the Future of Life Institute are international partners in the forum.

Between the lines: - Korea is positioning SFASS 2026 as a global convening point for AI safety, marking a second straight year in a leading role. - The participation of major U.S. tech companies signals that frontier model safety work is moving closer to practical deployment risks, not just benchmark testing. - The humanoid robot challenge suggests organizers want to stress-test AI systems under conditions that mirror real-world embodied use. - CEO Sangyoon Yu said real-world red-teaming is becoming essential as AI shifts toward autonomous agents and humanoids.

What's next: - AIM Intelligence and its global partners are expected to use the workshop to gather empirical data on model robustness and attack pathways. - The forum could shape future AI safety practices by linking research, red-teaming and deployment-oriented testing. - SFASS 2026 may further strengthen Korea’s role in international AI safety discussions.

The bottom line: - AIM Intelligence is turning AI safety into a live, hands-on test at one of the year’s most visible frontier AI forums.