https://ai45.shlab.org.cn/research/posts/ Recent content in Posts on AI45 Research https://ai45.shlab.org.cn/research/images/papermod-cover.png https://ai45.shlab.org.cn/research/images/papermod-cover.png Hugo -- 0.147.9 en PaperMod Contributors Tue, 10 Feb 2026 00:00:00 +0000 https://ai45.shlab.org.cn/research/posts/safeverse%E5%85%AC%E4%BC%97%E5%8F%B7%E6%8E%A8%E9%80%81-en/ Tue, 10 Feb 2026 00:00:00 +0000 https://ai45.shlab.org.cn/research/posts/safeverse%E5%85%AC%E4%BC%97%E5%8F%B7%E6%8E%A8%E9%80%81-en/ Transform a regular video into an interactive, physics-consistent 3D digital twin in minutes—supporting red-teaming, adversarial training, and online agent evolution. Open source. https://ai45.shlab.org.cn/research/posts/blog_test/ Fri, 16 Jan 2026 10:00:00 +0800 https://ai45.shlab.org.cn/research/posts/blog_test/ Exploring how Looped Transformers might bridge the gap between language outputs and subconscious representations via internal recurrence. https://ai45.shlab.org.cn/research/posts/safework-f1/ Mon, 21 Jul 2025 10:00:00 +0800 https://ai45.shlab.org.cn/research/posts/safework-f1/ From Vague Notions to Precise Coordinates: Mapping Long-Term AI Risks We Can’t Afford to Overlook https://ai45.shlab.org.cn/research/posts/safework-t1/ Mon, 21 Jul 2025 10:00:00 +0800 https://ai45.shlab.org.cn/research/posts/safework-t1/ A safety reasoning training platform designed for multimodal large models, dedicated to boosting training efficiency, ensuring trustworthy reasoning, and enabling seamless cross-domain adaptation. https://ai45.shlab.org.cn/research/posts/%E6%BD%9C%E4%BC%8F%E5%9C%A8%E4%BD%A0%E6%89%8B%E6%9C%BA%E9%87%8C%E7%9A%84%E6%99%BA%E8%83%BD%E7%89%B9%E5%B7%A5%E7%AB%AF%E4%BE%A7agent%E9%9C%80%E8%A6%81%E6%80%8E%E6%A0%B7%E7%9A%84%E5%AE%89%E5%85%A8%E5%B9%B4%E6%A3%80/ Fri, 18 Jul 2025 10:00:00 +0800 https://ai45.shlab.org.cn/research/posts/%E6%BD%9C%E4%BC%8F%E5%9C%A8%E4%BD%A0%E6%89%8B%E6%9C%BA%E9%87%8C%E7%9A%84%E6%99%BA%E8%83%BD%E7%89%B9%E5%B7%A5%E7%AB%AF%E4%BE%A7agent%E9%9C%80%E8%A6%81%E6%80%8E%E6%A0%B7%E7%9A%84%E5%AE%89%E5%85%A8%E5%B9%B4%E6%A3%80/ Building a high-fidelity, all-around security evaluation framework for mobile AI assistants, fortifying the security defense line for mobile phones to become users' 'second brain'. https://ai45.shlab.org.cn/research/posts/safework_e1/ Mon, 14 Jul 2025 10:00:00 +0800 https://ai45.shlab.org.cn/research/posts/safework_e1/ In response to the safety and trustworthiness challenges in the intelligent transformation of clean energy, Shanghai AI Lab, in collaboration with leading industry enterprises, has jointly created the first safety reasoning foundation model in this field. https://ai45.shlab.org.cn/research/posts/safework-r1/ Sat, 12 Jul 2025 09:00:00 +0800 https://ai45.shlab.org.cn/research/posts/safework-r1/ An advanced multimodal reasoning model, developed under our SafeLadder framework, that demonstrates the coevolution of capabilities and safety. https://ai45.shlab.org.cn/research/posts/safework-v1/ Sat, 12 Jul 2025 09:00:00 +0800 https://ai45.shlab.org.cn/research/posts/safework-v1/ An autoformalization verification system that builds mathematically provable AI safety guarantee system.