This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.
The open-sourcing of large language models (LLMs) accelerates application development, innovation, and scientific progress. This includes both base models, which are pre-trained on extensive datasets without alignment, and aligned models, deliberately designed to align with ethical standards and human values. Contrary to the prevalent assumption that the inherent instruction-following limitations of base LLMs serve as a safeguard against misuse, our investigation exposes a critical oversight in this belief. By deploying carefully designed demonstrations, our research demonstrates that base LLMs could effectively interpret and execute malicious instructions. To systematically assess these risks, we introduce a novel set of risk evaluation metrics. Empirical results reveal that the outputs from base LLMs can exhibit risk levels on par with those of models fine-tuned for malicious purposes. This vulnerability, requiring neither specialized knowledge nor training, can be manipulated by almost anyone, highlighting the substantial risk and the critical need for immediate attention to the base LLMs' security protocols.
Independent evaluation and red teaming are critical for identifying the risks posed by generative AI systems. However, the terms of service and enforcement strategies used by prominent AI companies to deter model misuse have disincentives on good faith safety evaluations. This causes some researchers to fear that conducting such research or releasing their findings will result in account suspensions or legal reprisal. Although some companies offer researcher access programs, they are an inadequate substitute for independent research access, as they have limited community representation, receive inadequate funding, and lack independence from corporate incentives. We propose that major AI developers commit to providing a legal and technical safe harbor, indemnifying public interest safety research and protecting it from the threat of account suspensions or legal reprisal. These proposals emerged from our collective experience conducting safety, privacy, and trustworthiness research on generative AI systems, where norms and incentives could be better aligned with public interests, without exacerbating model misuse. We believe these commitments are a necessary step towards more inclusive and unimpeded community efforts to tackle the risks of generative AI.
In environments where RGB images are inadequate, pressure maps is a viable alternative, garnering scholarly attention. This study introduces a novel self-supervised pressure map keypoint detection (SPMKD) method, addressing the current gap in specialized designs for human keypoint extraction from pressure maps. Central to our contribution is the Encoder-Fuser-Decoder (EFD) model, which is a robust framework that integrates a lightweight encoder for precise human keypoint detection, a fuser for efficient gradient propagation, and a decoder that transforms human keypoints into reconstructed pressure maps. This structure is further enhanced by the Classification-to-Regression Weight Transfer (CRWT) method, which fine-tunes accuracy through initial classification task training. This innovation not only enhances human keypoint generalization without manual annotations but also showcases remarkable efficiency and generalization, evidenced by a reduction to only $5.96\%$ in FLOPs and $1.11\%$ in parameter count compared to the baseline methods.
Backdoor attacks are commonly executed by contaminating training data, such that a trigger can activate predetermined harmful effects during the test phase. In this work, we present AnyDoor, a test-time backdoor attack against multimodal large language models (MLLMs), which involves injecting the backdoor into the textual modality using adversarial test images (sharing the same universal perturbation), without requiring access to or modification of the training data. AnyDoor employs similar techniques used in universal adversarial attacks, but distinguishes itself by its ability to decouple the timing of setup and activation of harmful effects. In our experiments, we validate the effectiveness of AnyDoor against popular MLLMs such as LLaVA-1.5, MiniGPT-4, InstructBLIP, and BLIP-2, as well as provide comprehensive ablation studies. Notably, because the backdoor is injected by a universal perturbation, AnyDoor can dynamically change its backdoor trigger prompts/harmful effects, exposing a new challenge for defending against backdoor attacks. Our project page is available at https://sail-sg.github.io/AnyDoor/.
Large language models (LLMs) are vulnerable to jailbreak attacks - resulting in harmful, unethical, or biased text generations. However, existing jailbreaking methods are computationally costly. In this paper, we propose the weak-to-strong jailbreaking attack, an efficient method to attack aligned LLMs to produce harmful text. Our key intuition is based on the observation that jailbroken and aligned models only differ in their initial decoding distributions. The weak-to-strong attack's key technical insight is using two smaller models (a safe and an unsafe one) to adversarially modify a significantly larger safe model's decoding probabilities. We evaluate the weak-to-strong attack on 5 diverse LLMs from 3 organizations. The results show our method can increase the misalignment rate to over 99% on two datasets with just one forward pass per example. Our study exposes an urgent safety issue that needs to be addressed when aligning LLMs. As an initial attempt, we propose a defense strategy to protect against such attacks, but creating more advanced defenses remains challenging. The code for replicating the method is available at https://github.com/XuandongZhao/weak-to-strong
The emergence of LLM-based agents has garnered considerable attention, yet their trustworthiness remains an under-explored area. As agents can directly interact with the physical environment, their reliability and safety is critical. This paper presents an Agent-Constitution-based agent framework, TrustAgent, an initial investigation into improving the safety dimension of trustworthiness in LLM-based agents. This framework consists of threefold strategies: pre-planning strategy which injects safety knowledge to the model prior to plan generation, in-planning strategy which bolsters safety during plan generation, and post-planning strategy which ensures safety by post-planning inspection. Through experimental analysis, we demonstrate how these approaches can effectively elevate an LLM agent's safety by identifying and preventing potential dangers. Furthermore, we explore the intricate relationships between safety and helpfulness, and between the model's reasoning ability and its efficacy as a safe agent. This paper underscores the imperative of integrating safety awareness and trustworthiness into the design and deployment of LLM-based agents, not only to enhance their performance but also to ensure their responsible integration into human-centric environments. Data and code are available at https://github.com/agiresearch/TrustAgent.
Large language models are meticulously aligned to be both helpful and harmless. However, recent research points to a potential overkill which means models may refuse to answer benign queries. In this paper, we investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to an over-attention of harmful words like 'kill' and prompts emphasizing safety will exacerbate overkill. Based on these insights, we introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon. We first extract such over-attention by amplifying the difference in the model's output distributions when responding to system prompts that either include or omit an emphasis on safety. Then we determine the final next-token predictions by downplaying the over-attention from the model via contrastive decoding. Empirical results indicate that our method has achieved an average reduction of the refusal rate by 20\% while having almost no impact on safety.
Large Language Models (LLMs) have exhibited remarkable capabilities in understanding and interacting with natural language across various sectors. However, their effectiveness is limited in specialized areas requiring high accuracy, such as plant science, due to a lack of specific expertise in these fields. This paper introduces PLLaMa, an open-source language model that evolved from LLaMa-2. It's enhanced with a comprehensive database, comprising more than 1.5 million scholarly articles in plant science. This development significantly enriches PLLaMa with extensive knowledge and proficiency in plant and agricultural sciences. Our initial tests, involving specific datasets related to plants and agriculture, show that PLLaMa substantially improves its understanding of plant science-related topics. Moreover, we have formed an international panel of professionals, including plant scientists, agricultural engineers, and plant breeders. This team plays a crucial role in verifying the accuracy of PLLaMa's responses to various academic inquiries, ensuring its effective and reliable application in the field. To support further research and development, we have made the model's checkpoints and source codes accessible to the scientific community. These resources are available for download at \url{https://github.com/Xianjun-Yang/PLLaMa}.