SafeWork-R1: Coevolving Safety and Intelligence under the AI-45$^{\circ}$ Law

TL;DR

SafeWork-R1 is a multimodal reasoning model that coevolves safety and intelligence through a novel SafeLadder framework, achieving significant safety improvements without sacrificing capabilities, and demonstrating the potential for trustworthy AI development.

Contribution

The paper introduces SafeLadder, a new framework enabling AI models to develop intrinsic safety reasoning and self-reflection, advancing safety-oriented reinforcement learning beyond existing methods.

Findings

SafeWork-R1 improves safety benchmark scores by 46.54% over base models.

It outperforms proprietary models like GPT-4.1 in safety performance.

The framework demonstrates safe and capable AI co-evolution across multiple models.

Abstract

We introduce SafeWork-R1, a cutting-edge multimodal reasoning model that demonstrates the coevolution of capabilities and safety. It is developed by our proposed SafeLadder framework, which incorporates large-scale, progressive, safety-oriented reinforcement learning post-training, supported by a suite of multi-principled verifiers. Unlike previous alignment methods such as RLHF that simply learn human preferences, SafeLadder enables SafeWork-R1 to develop intrinsic safety reasoning and self-reflection abilities, giving rise to safety `aha' moments. Notably, SafeWork-R1 achieves an average improvement of $46.54\%$ over its base model Qwen2.5-VL-72B on safety-related benchmarks without compromising general capabilities, and delivers state-of-the-art safety performance compared to leading proprietary models such as GPT-4.1 and Claude Opus 4. To further bolster its reliability, we implement two distinct inference-time intervention methods and a deliberative search mechanism, enforcing step-level verification. Finally, we further develop SafeWork-R1-InternVL3-78B, SafeWork-R1-DeepSeek-70B, and SafeWork-R1-Qwen2.5VL-7B. All resulting models demonstrate that safety and capability can co-evolve synergistically, highlighting the generalizability of our framework in building robust, reliable, and trustworthy general-purpose AI.