Red Teaming Visual Language Models

Anonymous

Abstract: VLMs (Vision-Language Models) extend the capabilities of LLMs (Large Language Models) to accept multimodal inputs. Since it has been verified that LLMs can be induced to generate harmful or inaccurate content through specific test cases (termed as Red Teaming), how VLMs perform in similar scenarios, especially with their combination of textual and visual inputs, remains a question. To explore this problem, we present a novel red teaming dataset RTVLM, which encompasses 12 subtasks (e.g., image misleading, multi-modal jailbreaking, face fairness, etc) under 4 primary aspects (faithfulness, privacy, safety, fairness). Our RTVLM is the first red teaming dataset to benchmark current VLMs in terms of these 4 different aspects. Detailed analysis shows that 10 prominent open-sourced VLMs struggle with the red teaming in different degrees and have up to 31% performance gap with GPT-4V. Additionally, we simply apply red teaming alignment to LLaVA-v1.5 with Supervised Fine-tuning (SFT) using RTVLM, and this bolsters the models' performance with 10% in RTVLM test set, 13% in MM-hallu, and without noticeable decline in MM-Bench, overpassing other LLaVA-based models in similar size with regular alignment data. This reveals that current open-sourced VLMs still lack red teaming alignment. Our code and datasets will be open-sourced.

Paper Type: long

Research Area: Multimodality and Language Grounding to Vision, Robotics and Beyond

Contribution Types: Data resources

Languages Studied: English

Preprint Status: There is a non-anonymous preprint (URL specified in the next question).

A1: yes

A1 Elaboration For Yes Or No: Section 6 Limitations

A2: yes

A2 Elaboration For Yes Or No: Section 7 Ethic consideration

A3: yes

A3 Elaboration For Yes Or No: Abstract and Introduction part

B: yes

B1: yes

B1 Elaboration For Yes Or No: Section 2 RTVLM Dataset

B2: yes

B2 Elaboration For Yes Or No: Section 2 RTVLM Dataset

B3: yes

B3 Elaboration For Yes Or No: Section 2 RTVLM Dataset

B4: yes

B4 Elaboration For Yes Or No: Section 2 RTVLM Dataset

B5: yes

B5 Elaboration For Yes Or No: Section 2 RTVLM Dataset

B6: yes

B6 Elaboration For Yes Or No: Section 2 RTVLM Dataset

C: no

C1: yes

C1 Elaboration For Yes Or No: Section 3 Experimental Results

C2: yes

C2 Elaboration For Yes Or No: 3.1 Experimental Settings

C3: yes

C3 Elaboration For Yes Or No: 3.2 Red Teaming Test Results

C4: yes

C4 Elaboration For Yes Or No: 3.2 Red Teaming Test Results

D: yes

D1: yes

D1 Elaboration For Yes Or No: Appendix D

D2: yes

D2 Elaboration For Yes Or No: Ethical consideration

D3: yes

D3 Elaboration For Yes Or No: Ethical consideration

D4: yes

D4 Elaboration For Yes Or No: Ethical consideration

D5: yes

D5 Elaboration For Yes Or No: Ethical consideration

E: yes

E1: yes

E1 Elaboration For Yes Or No: Ethical consideration