REVIEW 3 major objections 6 minor 83 references
Evaluating the efficacy of LLM Safety Solutions : The Palit Benchmark Dataset
T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read The paper builds a 500-prompt benchmark of malicious and benign prompts, evaluates seven LLM safety filters and a ChatGPT-3.5-Turbo baseline, and reports that the baseline's false-positive rate (0.438–0.859) makes it unusable as a safety…
desk verdict Useful first cut at benchmarking commercial LLM security tools, but the headline rankings are not load-bearing on their own: the benchmark isn't released and the results are prompt-source-dependent. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the benchmark itself: a collection of 500 prompts built from hand-written attacks and the outputs of three automated attack generators, paired with a benign-prompt set and an existing public dataset for cross-checking. Prompts are grounded in four application scenarios—French translation, a finance chatbot, a code-to-SQL tool, and a general-purpose chat agent—and are tested with and without the system prompt to isolate context sensitivity. The argument runs through standard detection metrics, including precision, recall, accuracy, false-positive rate, F1 score, latency, and attack success rate, so the tool ranking is an empirical comparison rather than a feature checklist.
What would settle it
Collect a corpus of malicious prompts actually observed against deployed LLM-integrated applications, for example from production logs or bug-bounty reports, run the same seven tools plus the baseline on it, and check whether Lakera Guard still has the lowest false-positive rate and highest F1 score and whether the baseline still exceeds a 0.4 false-positive rate; a substantial reordering would show the benchmark's results are artifacts of its synthetic prompt mix.
Extended reading notes
Core claim
On its own terms, the paper claims that the emerging LLM safety-tool market has no obvious winner and that simple comparisons are misleading. The Palit benchmark evaluates seven filters—four open-source and three closed-source—plus ChatGPT-3.5-Turbo as a baseline. Lakera Guard posts the strongest combination across tests, with F1 scores of 0.659 without context, 0.809 with context, and 0.823 on an existing public prompt-injection dataset, along with sub-0.1-second latency; ProtectAI LLM Guard is close in precision but an order of magnitude slower. The baseline model often appears accurate because the dataset is imbalanced, yet its false-positive rates of 0.438–0.859 make it impractical. The paper also claims that the source of prompts matters: manually crafted attacks are the hardest to stop, while the behavior of generator-produced attacks changes sharply when context is added.
Load-bearing premise
The benchmark's malicious prompts are generated by researchers using manual techniques and automated attack tools, and the paper concedes it is unclear whether these resemble the attacks real adversaries would actually send to production LLM applications; if real-world attack distribution differs, the rankings may not transfer.
Editorial extensions
If this is right
- Relying on a foundation model's built-in safety is not enough: at false-positive rates between 0.438 and 0.859, a production chatbot would flag or block a large share of legitimate requests.
- If a deployer values both detection and low disruption, the results favor Lakera Guard and ProtectAI LLM Guard over the other five tested tools.
- Context is a first-order variable: adding system prompts changed attack success rates sharply, including pushing manual prompts to 100% success against the baseline, so evaluations need application context to be meaningful.
- Open-source tools in the study were hampered by outdated code and long latencies—one tool timed out on longer prompts and took seconds to minutes per prompt—so maintenance and speed are part of the safety story.
- Accuracy is misleading on imbalanced attack/benign splits; false-positive rate and attack success rate should be reported alongside accuracy to judge a filter.
Reading between the lines
- A testable extension would be to rerun the same seven tools on a corpus of attacks observed in real deployments, for example from production logs or bug-bounty reports; if the ranking changes, the synthetic prompt mix is not representative of field conditions.
- The across-the-board improvement on the public prompt-injection dataset raises the possibility that several commercial filters have been trained or tuned on public benchmark data, which would make public-dataset results an upper bound rather than a field estimate.
- The wide latency spread, from about 0.035 to over 29 seconds per prompt, implies an operational cost the paper leaves implicit: at production query volumes, latency alone may decide tool adoption even when a slower tool has marginally better detection.
- The context ablation suggests that filter deployment should be tuned against the target application's actual prompt template, not validated in isolation.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper presents a comparative evaluation of seven LLM security tools (plus a ChatGPT-3.5-Turbo baseline) against a newly constructed benchmark called Palit, comprising 500 prompts (75.8% malicious) drawn from manual generation, Houyi, Garak, PromptMap, and the public Deepset prompt-injection dataset. The authors report precision, recall, accuracy, FPR, F1, latency, and attack success rates under three conditions: no context, with base-prompt context, and on the Deepset dataset. Their main conclusions are that the ChatGPT-3.5-Turbo baseline is unsuitable due to high false-positive rates (FPR 0.438-0.859), that Lakera Guard and ProtectAI LLM Guard are the best overall tools, and that the market of bolt-on LLM security solutions suffers from limited transparency and usability issues. The paper also surveys 13 tools and compares their claimed features.
Significance. If the results hold, this would be one of the first independent, published evaluations of commercial and open-source LLM security tools, and it would provide a useful template for future benchmarking in this area. The paper's strengths are its timely research question, the explicit attention to usability (latency, deployment, trial availability), the multi-source attack generation strategy, the ablation on system-prompt context, and the remarkably honest limitations section. The data tables support the qualitative claim that the baseline has a very high FPR and that several tools (especially Lakera, ProtectAI, and Azure Prompt Shield) achieve high precision with low FPR. However, the headline 'best overall' ranking is built on an undefined composite metric and on a benchmark whose representativeness of real-world attacks the authors themselves disclaim; the per-source ASR tables show that the ranking shifts substantially depending on prompt source. The paper is therefore a useful empirical contribution whose central claim needs substantial qualification before it can be accepted as stated.
major comments (3)
- [Sections 4.3 and 5.2] The claim that 'Lakera Guard and ProtectAI LLM Guard emerged as the best overall tools' is not supported by any explicit definition of 'best overall.' The paper reports six separate metrics (accuracy, precision, recall, FPR, F1, latency) and no rule for combining them. Different tools win on different dimensions: e.g., Lakera has the highest accuracy on the benchmark with context (74.6%) and lowest latency (0.305s), but Azure Prompt Shield has higher precision (0.952) and lower FPR (0.074); on the Deepset dataset, Azure achieves precision 1.0 while Lakera achieves higher recall. Without a stated aggregation rule or a decision-theoretic justification (e.g., a cost model for false positives vs. false negatives vs. latency), the ordering is not reproducible and the 'best overall' wording should be replaced with a statement about specific metrics.
- [Section 5.3, Tables 6 and 7] The external validity of the Palit benchmark is load-bearing for the ranking, and the manuscript itself disclaims it: 'it is unclear whether our attacks are representative of malicious prompts used by real-world attackers, given they were inspired by attacks created by researchers and not threat intelligence.' The internal results show that the ranking is not invariant across attack sources. For example, with context the baseline ASR is 100% on manual prompts but 0% on Garak and PromptMap prompts (Table 7), and Lakera's ASR on PromptMap prompts is 83.05% even though its overall ASR is low. This means the headline ranking could be an artifact of the particular 75.8%-malicious mix of manual/Houyi/Garak/PromptMap prompts. The authors should either (a) provide evidence that this mix approximates real-world malicious prompt distributions, or (b) restrict their conclusions to per-source results and characterize for which prompt families each tool is effective. Since the dataset is not released (no link or repository is given), readers cannot even check for overlaps with vendor training data, a risk the authors themselves raise for the Deepset portion in Section 5.2.
- [Section 3.3 and Tables 3-7] All reported metrics are point estimates on a single dataset of 500 prompts, with no confidence intervals, bootstrap estimates, or significance tests. This is especially problematic for the per-source ASR tables, where the number of prompts per cell is small (e.g., the Garak-generated subset contains 13 DAN prompts in addition to encoding-based attacks, and the manual subset yields only 59-81 prompts per context condition). Differences that appear large, such as Vigil's drop in ASR from 93.22% to 70.37% on PromptMap prompts when context is added, may be within sampling noise. The authors should report uncertainties or at least explicitly discuss the statistical power limits before drawing conclusions about comparative performance.
minor comments (6)
- [Abstract and Section 1] Grammar: 'evaluate these tools performance' should be 'evaluate these tools' performance'; 'addressed the following research questions' mixes tenses. The abstract would also benefit from stating the number of tools evaluated and the dataset size explicitly.
- [Section 2.5] Typo: 'Similarlym, Greshake et al.' should be 'Similarly, Greshake et al.'
- [Section 3.2] The benchmark dataset is not made available: no link, repository, or data availability statement is provided. For a paper whose main contribution is a 'benchmark dataset,' releasing it (or an anonymized subset) is essential for reproducibility and for verifying the authors' own concern about training-data overlap.
- [Tables 6 and 7] The column header 'PAI Guard' should be 'ProtectAI LLM Guard' for consistency with Tables 3-5. Also, 'LK Similarity' and 'LK Canary' should be expanded or defined in the caption, since the abbreviations are not introduced in the table.
- [Reference [79]] Reference [79] in Section 3.2 is the same paper as [43] (Liu et al., 'Prompt Injection attack against LLM-integrated Applications'); duplicate citations should be merged.
- [Table 2] The table contains formatting issues, including a cut-off '/times✓' in the Output Scan. row and the header 'PAI Guard' (should be 'ProtectAI LLM Guard'). The legend says '-' for 'no information available,' but the cell for 'Jailbreak' in the ArthurAI column shows '-' which is inconsistent with the claim in the text that 'Arthur AI Firewall' provides jailbreak protection.
Circularity Check
No circularity found: the paper is a direct empirical benchmark evaluation whose conclusions are computed from measured tool outputs, with no fitted inputs renamed as predictions and no load-bearing self-citation chain.
full rationale
This paper makes no derivation claims of the kind that can be circular. Its central results—relative tool rankings, false positive rates, attack success rates, and latency—are obtained by running seven security tools and a baseline model against a newly constructed benchmark dataset and reporting standard metrics (precision, recall, accuracy, FPR, F1, latency, ASR). The benchmark prompts were generated manually and with third-party generators (Houyi, Garak, PromptMap) plus the public Deepset dataset, independently of the tool outputs; no parameter was fitted to a subset of the evaluation data and then renamed as a prediction. The paper does not rest on a self-citation: the authors cite prior work for background, attack taxonomies, and tool descriptions, but the evaluative conclusions are supported by the measurements reported in Tables 3-7. The two validity concerns the authors themselves flag—whether research-generated prompts represent real-world attacks (Section 5.3) and the possibility that some tools were trained on the Deepset dataset (Section 5.2)—are external-validity and data-contamination risks, not circular reasoning. Acknowledging a limitation about representativeness does not make the benchmark's definition depend on the conclusion being drawn. Therefore, under the stated rules, the appropriate finding is no significant circularity with score 0.
Assumptions & free parameters
assumptions (4)
- domain assumption The 500-prompt Palit benchmark, with 75.8% malicious prompts, is a valid instrument for measuring tool effectiveness.
- domain assumption Binary classification by each tool (malicious or benign) via API responses is an accurate measure of the tool's security capability.
- domain assumption The sandwich defense prompt used for the ChatGPT-3.5-Turbo baseline ensures the baseline is classifying rather than following the user prompt.
- domain assumption The publicly available Deepset dataset labels are correct ground truth for maliciousness.
Cite this review
Pith. "Pith review of Evaluating the efficacy of LLM Safety Solutions : The Palit Benchmark Dataset." pith.science (2026). https://pith.science/paper/OC3IC6WI
@misc{pith2026250513028,
author = {Pith},
title = {Pith review of: Evaluating the efficacy of LLM Safety Solutions : The Palit Benchmark Dataset},
year = {2026},
howpublished = {\url{https://pith.science/paper/OC3IC6WI}},
note = {Machine review of arXiv:2505.13028}
}
read the original abstract
Large Language Models (LLMs) are increasingly integrated into critical systems in industries like healthcare and finance. Users can often submit queries to LLM-enabled chatbots, some of which can enrich responses with information retrieved from internal databases storing sensitive data. This gives rise to a range of attacks in which a user submits a malicious query and the LLM-system outputs a response that creates harm to the owner, such as leaking internal data or creating legal liability by harming a third-party. While security tools are being developed to counter these threats, there is little formal evaluation of their effectiveness and usability. This study addresses this gap by conducting a thorough comparative analysis of LLM security tools. We identified 13 solutions (9 closed-source, 4 open-source), but only 7 were evaluated due to a lack of participation by proprietary model owners.To evaluate, we built a benchmark dataset of malicious prompts, and evaluate these tools performance against a baseline LLM model (ChatGPT-3.5-Turbo). Our results show that the baseline model has too many false positives to be used for this task. Lakera Guard and ProtectAI LLM Guard emerged as the best overall tools showcasing the tradeoff between usability and performance. The study concluded with recommendations for greater transparency among closed source providers, improved context-aware detections, enhanced open-source engagement, increased user awareness, and the adoption of more representative performance metrics.
Figures
Reference graph
Works this paper leans on
-
[1]
https://github.com
Github. https://github.com. Accessed: 2024-08-27. 3.1
2024
-
[2]
https://scholar.google.com
Google scholar. https://scholar.google.com. Accessed: 2024-08-27. 3.1
2024
-
[3]
https://simonwillison.net/ 2024/Mar/5/prompt-injection-jailbreaking/
Prompt injection and jailbreaking are not the same thing — simonwillison.net. https://simonwillison.net/ 2024/Mar/5/prompt-injection-jailbreaking/ . [Accessed 24-08-2024]. 2.3
2024
- [4]
-
[5]
The vector database to build knowledgeable AI | Pinecone — pinecone.io. https://www.pinecone.io/. [Accessed 25-08-2024]. 3.3
work page 2024
- [6]
-
[7]
https://learnprompting.org/docs/prompt_hacking/defensive_measures/ sandwich_defense, 2023
Sandwitch defense. https://learnprompting.org/docs/prompt_hacking/defensive_measures/ sandwich_defense, 2023. 2.4, 3.3
work page 2023
-
[8]
Conversational health agents: A personalized llm-powered agent framework
Mahyar Abbasian, Iman Azimi, Amir M Rahmani, and Ramesh Jain. Conversational health agents: A personalized llm-powered agent framework. arXiv preprint arXiv:2310.02374, 2023. 1
arXiv 2023
Show all 83 references
-
[9]
Gpt-4 technical report
Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. 1
2023 arXiv
-
[10]
Vulnerabilities in personal firewalls caused by poor security usability
Bander Alfayyadh, James Ponting, Mohammed Alzomai, and Audun Jøsang. Vulnerabilities in personal firewalls caused by poor security usability. In 2010 IEEE International Conference on Information Theory and Information Security, pages 682–688. IEEE, 2010. 2.5
2010
-
[11]
Real Attackers Don’t Compute Gradients
Giovanni Apruzzese, Hyrum S. Anderson, Savino Dambra, David Freeman, Fabio Pierazzi, and Kevin Roundy. “Real Attackers Don’t Compute Gradients”: Bridging the Gap Between Adversarial ML Research and Practice. In 2023 IEEE Conference on Secure and Trustworthy Machine Learning (S...
2023
-
[12]
Protection — arthur.ai
ArthurAI. Protection — arthur.ai. https://www.arthur.ai/solution/firewall. [Accessed 26-08-2024]. 4.1
2024
-
[13]
LLM Hacking: Prompt Injection Techniques, July 2023
Austin Stubbs. LLM Hacking: Prompt Injection Techniques, July 2023. 2.2, 3.2
2023
-
[14]
Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Nee- lakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Je...
2020
-
[15]
A LLM Assisted Exploitation of AI-Guardian, July 2023
Nicholas Carlini. A LLM Assisted Exploitation of AI-Guardian, July 2023. arXiv:2307.15008 [cs]. 2.1, 5.4
2023 arXiv
-
[16]
Vigil | Vigil: Documentation — vigil.deadbits.ai
DeadBitsAI. Vigil | Vigil: Documentation — vigil.deadbits.ai. https://vigil.deadbits.ai/. [Accessed 26-08-2024]. 4.1
2024
-
[17]
Gemini by google deepmind
Google DeepMind. Gemini by google deepmind. Accessed: 2024-08-26, 2023. https://www.deepmind.com/ gemini. 2.1
2024
-
[18]
deepset/prompt-injections · Datasets at Hugging Face — huggingface.co
deepset.ai. deepset/prompt-injections · Datasets at Hugging Face — huggingface.co. https://huggingface. co/datasets/deepset/prompt-injections. [Accessed 25-08-2024]. 3.2, 3.2, 4.3
2024
-
[19]
garak: A Framework for Security Probing Large Language Models, June 2024
Leon Derczynski, Erick Galinkin, Jeffrey Martin, Subho Majumdar, and Nanna Inie. garak: A Framework for Security Probing Large Language Models, June 2024. arXiv:2406.11036 [cs]. 2.5, 3.1, 3.2, 3.3
2024 arXiv
-
[20]
Machine learning models predicting returns: Why most popular performance metrics are misleading and proposal for an efficient metric
Jean Dessain. Machine learning models predicting returns: Why most popular performance metrics are misleading and proposal for an efficient metric. Expert Systems with Applications, 199:116970, 2022. 5.2
2022
-
[21]
Bert: Pre-training of deep bidirectional transformers for language understanding
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langu...
2019
-
[22]
How should pre-trained language models be fine-tuned towards adversarial robustness? Advances in Neural Information Processing Systems , 34:4356–4369, 2021
Xinshuai Dong, Anh Tuan Luu, Min Lin, Shuicheng Yan, and Hanwang Zhang. How should pre-trained language models be fine-tuned towards adversarial robustness? Advances in Neural Information Processing Systems , 34:4356–4369, 2021. 2.4
2021
-
[23]
Comparing sql injection detection tools using attack injection: An experimental study
Ivano Alessandro Elia, Jose Fonseca, and Marco Vieira. Comparing sql injection detection tools using attack injection: An experimental study. In 2010 IEEE 21st International Symposium on Software Reliability Engineering, pages 289–298. IEEE, 2010. 2.5
2010
-
[24]
Comparing sql injection detection tools using attack injection: An experimental study
Ivano Alessandro Elia, Jose Fonseca, and Marco Vieira. Comparing sql injection detection tools using attack injection: An experimental study. In 2010 IEEE 21st International Symposium on Software Reliability Engineering, pages 289–298, 2010. 3.3
2010
-
[25]
Testing and comparing web vulnerability scanning tools for sql injection and xss attacks
Jose Fonseca, Marco Vieira, and Henrique Madeira. Testing and comparing web vulnerability scanning tools for sql injection and xss attacks. In 13th Pacific Rim international symposium on dependable computing (PRDC 2007), pages 365–372. IEEE, 2007. 2.5
2007
-
[26]
Testing and comparing web vulnerability scanning tools for sql injection and xss attacks
Jose Fonseca, Marco Vieira, and Henrique Madeira. Testing and comparing web vulnerability scanning tools for sql injection and xss attacks. In 13th Pacific Rim International Symposium on Dependable Computing (PRDC 2007), pages 365–372, 2007. 3.3
2007
-
[27]
Challenges in the real world use of classification accuracy metrics: From recall and precision to the matthews correlation coefficient
Giles M Foody. Challenges in the real world use of classification accuracy metrics: From recall and precision to the matthews correlation coefficient. Plos one, 18(10):e0291908, 2023. 5.2
2023
-
[28]
Github copilot
GitHub. Github copilot. Accessed: 2024-08-26, 2021. https://github.com/features/copilot. 2.1
2024
-
[29]
Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection, May 2023
Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz. Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection, May 2023. arXiv:2302.12173 [cs]. 2.5
2023 arXiv
-
[30]
Benchmarking approach to compare web applications static analysis tools detecting owasp top ten security vulnerabilities
Juan R Bermejo Higuera, Javier Bermejo Higuera, Juan A Sicilia Montalvo, Javier Cubo Villalba, and Juan José Nombela Pérez. Benchmarking approach to compare web applications static analysis tools detecting owasp top ten security vulnerabilities. Computers, Materials & Continua...
2020
-
[31]
Summon a Demon and Bind it: A Grounded Theory of LLM Red Teaming in the Wild, November 2023
Nanna Inie, Jonathan Stray, and Leon Derczynski. Summon a Demon and Bind it: A Grounded Theory of LLM Red Teaming in the Wild, November 2023. arXiv:2311.06237 [cs]. 2.5
2023 arXiv
-
[32]
Protect your AI applications in real time — Robust Intelligence — robustintelligence.com
Robust Intelligence. Protect your AI applications in real time — Robust Intelligence — robustintelligence.com. https://www.robustintelligence.com/platform/ai-firewall-guardrails . [Accessed 26-08-2024]. 4.1
2024
-
[33]
Chatgpt for good? on opportunities and challenges of large language models for education
Enkelejda Kasneci, Kathrin Seßler, Stefan Küchemann, Maria Bannert, Daryna Dementieva, Frank Fischer, Urs Gasser, Georg Groh, Stephan Günnemann, Eyke Hüllermeier, et al. Chatgpt for good? on opportunities and challenges of large language models for education. Learning and indi...
2023
-
[34]
Software updates as a security metric: Passive identification of update trends and effect on machine infection
Moazzam Khan, Zehui Bi, and John A Copeland. Software updates as a security metric: Passive identification of update trends and effect on machine infection. In MILCOM 2012-2012 IEEE Military Communications Conference, pages 1–6. IEEE, 2012. 5.1 15 arXiv Template A PREPRINT
2012
-
[35]
i have no idea what i’m doing
Katharina Krombholz, Wilfried Mayer, Martin Schmiedecker, and Edgar Weippl. " i have no idea what i’m doing"-on the usability of deploying{HTTPS}. In 26th USENIX Security Symposium (USENIX Security 17) , pages 1339–1356, 2017. 2.5
2017
-
[36]
Watch your language: large language models and content moderation
Deepak Kumar, Yousef AbuHashem, and Zakir Durumeric. Watch your language: large language models and content moderation. arXiv preprint arXiv:2309.14517, 2023. 1
2023 arXiv
-
[37]
Cummings, and Alexander Stimpson
Surender Suresh Kumar, M.L. Cummings, and Alexander Stimpson. Strengthening LLM Trust Boundaries: A Survey of Prompt Injection Attacks Surender Suresh Kumar Dr. M.L. Cummings Dr. Alexander Stimpson. In 2024 IEEE 4th International Conference on Human-Machine Systems (ICHMS), pa...
2024
-
[38]
Weight poisoning attacks on pre-trained models
Keita Kurita, Paul Michel, and Graham Neubig. Weight poisoning attacks on pre-trained models. arXiv preprint arXiv:2004.06660, 2020. 2.1
2004 arXiv
-
[39]
12 Top LLM Security Tools: Paid & Free (Overview) | Lakera – Protecting AI teams that disrupt the world
Lakera. 12 Top LLM Security Tools: Paid & Free (Overview) | Lakera – Protecting AI teams that disrupt the world. 1, 3.1
-
[40]
End-to-End Security for the Generative AI Era — lasso.security
Lasso. End-to-End Security for the Generative AI Era — lasso.security. https://www.lasso.security/. [Accessed 26-08-2024]. 4.1
2024
-
[41]
Learn Prompting: Your Guide to Communicating with AI
Learn Prompting. Learn Prompting: Your Guide to Communicating with AI. 3.1
-
[42]
AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models, March 2024
Xiaogeng Liu, Nan Xu, Muhao Chen, and Chaowei Xiao. AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models, March 2024. arXiv:2310.04451 [cs]. 2.5
2024 arXiv
-
[44]
Formalizing and benchmarking prompt injection attacks and defenses, 2024
Yupei Liu, Yuqi Jia, Runpeng Geng, Jinyuan Jia, and Neil Zhenqiang Gong. Formalizing and benchmarking prompt injection attacks and defenses, 2024. 2.2, 2.3, 2.5, 3.2
2024
-
[45]
Benchmarking of machine learning for anomaly based intrusion detection systems in the cicids2017 dataset
Ziadoon Kamil Maseer, Robiah Yusof, Nazrulazhar Bahaman, Salama A Mostafa, and Cik Feresa Mohd Foozy. Benchmarking of machine learning for anomaly based intrusion detection systems in the cicids2017 dataset. IEEE access, 9:22351–22370, 2021. 3.3
2021
-
[46]
PurpleLlama/Prompt-Guard/MODEL_card.md at main · meta-llama/PurpleLlama
Meta. PurpleLlama/Prompt-Guard/MODEL_card.md at main · meta-llama/PurpleLlama. 2.4, 3.1
-
[47]
Introducing ChatGPT
OpenAI. Introducing ChatGPT. https://openai.com/blog/chatgpt, 2023. 3.3
2023
-
[48]
Training language models to follow instructions with human feedback
Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:2773...
2022
-
[49]
LLM Top 10 for LLMs v1.1
OW ASP. LLM Top 10 for LLMs v1.1. 2, 2.5, 3.1, 3.2
-
[50]
Comparative analysis of commercial and open source mobile device forensic tools
Radhika Padmanabhan, Karen Lobo, Mrunali Ghelani, Dhanika Sujan, and Mahesh Shirole. Comparative analysis of commercial and open source mobile device forensic tools. In 2016 Ninth International Conference on Contemporary Computing (IC3), pages 1–6. IEEE, 2016. 2.5
2016
-
[51]
Prompt Shields in Azure AI Content Safety - Azure AI services — learn.microsoft.com
PatrickFarley. Prompt Shields in Azure AI Content Safety - Azure AI services — learn.microsoft.com. https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/ jailbreak-detection. [Accessed 25-08-2024]. 3.3, 4.1
2024
-
[52]
Ignore Previous Prompt: Attack Techniques For Language Models, November 2022
Fábio Perez and Ian Ribeiro. Ignore Previous Prompt: Attack Techniques For Language Models, November 2022. arXiv:2211.09527 [cs]. 2.2, 2.5
2022 arXiv
-
[53]
Protect AI
Inc. Protect AI. Index - LLM Guard — llm-guard.com. https://llm-guard.com/. [Accessed 26-08-2024]. 4.1
2024
-
[54]
GitHub - protectai/rebuff: LLM Prompt Injection Detector — github.com
ProtectAI. GitHub - protectai/rebuff: LLM Prompt Injection Detector — github.com. https://github.com/ protectai/rebuff. [Accessed 25-08-2024]. 3.3, 4.1
2024
-
[55]
Fine-tuned deberta-v3 for prompt injection detection, 2023
ProtectAI.com. Fine-tuned deberta-v3 for prompt injection detection, 2023. 2.4, 3.1
2023
-
[56]
Software vulnerability detection using large language models
Moumita Das Purba, Arpita Ghosh, Benjamin J Radford, and Bill Chu. Software vulnerability detection using large language models. In 2023 IEEE 34th International Symposium on Software Reliability Engineering Workshops (ISSREW), pages 112–119. IEEE, 2023. 1
2023
-
[57]
An Early Catego- rization of Prompt Injection Attacks on Large Language Models, January 2024
Sippo Rossi, Alisia Marianne Michel, Raghava Rao Mukkamala, and Jason Bennett Thatcher. An Early Catego- rization of Prompt Injection Attacks on Large Language Models, January 2024. arXiv:2402.00898 [cs]. 2.2, 2.5, 3.2 16 arXiv Template A PREPRINT
2024 arXiv
-
[58]
JasperLS/gelectra-base-injection· Hugging Face — huggingface.co
Jasper Schwenzow. JasperLS/gelectra-base-injection· Hugging Face — huggingface.co. https://huggingface. co/JasperLS/gelectra-base-injection . [Accessed 25-08-2024]. 4.1
2024
-
[59]
Aim | AI-FIREWALL — aim.security
AIM Security. Aim | AI-FIREWALL — aim.security. https://www.aim.security/solutions/ aim-ai-firewall. [Accessed 26-08-2024]. 4.1
2024
-
[60]
LLM Security — llmsecurity.net
LLM Security. LLM Security — llmsecurity.net. https://llmsecurity.net/. [Accessed 26-08-2024]. 3.1
2024
-
[61]
Prompt Security: The Platform for GenAI Security — prompt.security
Prompt Security. Prompt Security: The Platform for GenAI Security — prompt.security. https://www.prompt. security/. [Accessed 26-08-2024]. 4.1
2024
-
[62]
GitHub - utkusen/promptmap: automatically tests prompt injection attacks on ChatGPT instances — github.com
Utku Sen. GitHub - utkusen/promptmap: automatically tests prompt injection attacks on ChatGPT instances — github.com. https://github.com/utkusen/promptmap. [Accessed 25-08-2024]. 3.2
2024
-
[63]
Poison frogs! targeted clean-label poisoning attacks on neural networks.Advances in neural information processing systems, 31, 2018
Ali Shafahi, W Ronny Huang, Mahyar Najibi, Octavian Suciu, Christoph Studer, Tudor Dumitras, and Tom Goldstein. Poison frogs! targeted clean-label poisoning attacks on neural networks.Advances in neural information processing systems, 31, 2018. 2.1
2018
-
[64]
Large language model alignment: A survey
Tianhao Shen, Renren Jin, Yufei Huang, Chuang Liu, Weilong Dong, Zishan Guo, Xinwei Wu, Yan Liu, and Deyi Xiong. Large language model alignment: A survey. arXiv preprint arXiv:2309.15025, 2023. 2.4
2023 arXiv
-
[65]
Do Anything Now
Xinyue Shen, Zeyuan Chen, Michael Backes, Yun Shen, and Yang Zhang. "Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models, May 2024. arXiv:2308.03825 [cs]. 2.3
2024 arXiv
-
[66]
Punctuation matters! stealthy backdoor attack for language models
Xuan Sheng, Zhicheng Li, Zhaoyang Han, Xiangmao Chang, and Piji Li. Punctuation matters! stealthy backdoor attack for language models. In CCF International Conference on Natural Language Processing and Chinese Computing, pages 524–536. Springer, 2023. 1, 2.1
2023
-
[67]
Performance comparison of intrusion detection machine learning classifiers on benchmark and new datasets
Mohammed F Suleiman and Biju Issac. Performance comparison of intrusion detection machine learning classifiers on benchmark and new datasets. In 2018 28th International Conference on Computer Theory and Applications (ICCTA), pages 19–23. IEEE, 2018. 3.3
2018
-
[68]
Aligning large multimodal models with factually augmented rlhf
Zhiqing Sun, Sheng Shen, Shengcao Cao, Haotian Liu, Chunyuan Li, Yikang Shen, Chuang Gan, Liang-Yan Gui, Yu-Xiong Wang, Yiming Yang, et al. Aligning large multimodal models with factually augmented rlhf. arXiv preprint arXiv:2309.14525, 2023. 2.4
2023 arXiv
-
[69]
Adversarial machine learning : a taxonomy and terminology of attacks and mitigations
Apostol Vassilev, Alina Oprea, Alie Fordyce, and Hyrum Anderson. Adversarial machine learning : a taxonomy and terminology of attacks and mitigations. Technical Report NIST 100-2e2023, National Institute of Standards and Technology (U.S.), Gaithersburg, MD, January 2024. 1, 2,...
2024
-
[70]
Attention is all you need
Ashish Vaswani. Attention is all you need. arXiv preprint arXiv:1706.03762, 2017. 1
2017 arXiv
-
[71]
Jailbroken: How does llm safety training fail? Advances in Neural Information Processing Systems, 36, 2024
Alexander Wei, Nika Haghtalab, and Jacob Steinhardt. Jailbroken: How does llm safety training fail? Advances in Neural Information Processing Systems, 36, 2024. 2.3
2024
-
[72]
Why johnny can’t encrypt: A usability evaluation of pgp 5.0
Alma Whitten and J Doug Tygar. Why johnny can’t encrypt: A usability evaluation of pgp 5.0. In USENIX security symposium, volume 348, pages 169–184, 1999. 2.5
1999
-
[73]
GitHub - whylabs/langkit: LangKit: An open-source toolkit for monitoring Large Language Models (LLMs)
WhyLabs. GitHub - whylabs/langkit: LangKit: An open-source toolkit for monitoring Large Language Models (LLMs). Extracts signals from prompts & responses, ensuring safety & security. Features include text quality, relevance metrics, & sentiment analysis. A comprehensive tool f...
2024
-
[74]
LLM Security Management — whylabs.ai
WhyLabs. LLM Security Management — whylabs.ai. https://whylabs.ai/llm-security. [Accessed 26-08-2024]. 4.1
2024
-
[75]
Bloomberggpt: A large language model for finance
Shijie Wu, Ozan Irsoy, Steven Lu, Vadim Dabravolski, Mark Dredze, Sebastian Gehrmann, Prabhanjan Kambadur, David Rosenberg, and Gideon Mann. Bloomberggpt: A large language model for finance. arXiv preprint arXiv:2303.17564, 2023. 1, 2.1
2023 arXiv
-
[76]
Watch out for your agents! investigating backdoor threats to llm-based agents
Wenkai Yang, Xiaohan Bi, Yankai Lin, Sishuo Chen, Jie Zhou, and Xu Sun. Watch out for your agents! investigating backdoor threats to llm-based agents. arXiv preprint arXiv:2402.11208, 2024. 1
2024 arXiv
-
[77]
Poisonprompt: Backdoor attack on prompt-based large language models
Hongwei Yao, Jian Lou, and Zhan Qin. Poisonprompt: Backdoor attack on prompt-based large language models. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 7745–7749. IEEE, 2024. 1, 2.1
2024
-
[78]
A survey on large language model (llm) security and privacy: The good, the bad, and the ugly
Yifan Yao, Jinhao Duan, Kaidi Xu, Yuanfang Cai, Zhibo Sun, and Yue Zhang. A survey on large language model (llm) security and privacy: The good, the bad, and the ugly. High-Confidence Computing, page 100211, 2024. 1, 2 17 arXiv Template A PREPRINT
2024
-
[79]
Prompt Injection attack against LLM-integrated Applications, March
Yi Liu, Gelei Deng, Yuekang Li, Kailong Wang, Zihao Wang, Xiaofeng Wang, Tianwei Zhang, Yepang Liu, Haoyu Wang, Yan Zheng, and Yang Liu. Prompt Injection attack against LLM-integrated Applications, March
-
[80]
Gptfuzzer: Red teaming large language models with auto-generated jailbreak prompts
Jiahao Yu, Xingwei Lin, and Xinyu Xing. Gptfuzzer: Red teaming large language models with auto-generated jailbreak prompts. arXiv preprint arXiv:2309.10253, 2023. 3.1
2023 arXiv
-
[81]
arXiv:2306.05499 [cs]. 3.2
-
[82]
Prompt as triggers for backdoor attack: Examining the vulnerability in language models
Shuai Zhao, Jinming Wen, Luu Anh Tuan, Junbo Zhao, and Jie Fu. Prompt as triggers for backdoor attack: Examining the vulnerability in language models. arXiv preprint arXiv:2305.01219, 2023. 1, 2.1
2023 arXiv
-
[83]
InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents, March 2024
Qiusi Zhan, Zhixiang Liang, Zifan Ying, and Daniel Kang. InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents, March 2024. arXiv:2403.02691 [cs]. 2.5
2024 arXiv
-
[85]
Zico Kolter, and Matt Fredrikson
Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, and Matt Fredrikson. Universal and Transferable Adversarial Attacks on Aligned Language Models, December 2023. arXiv:2307.15043 [cs]. 2.1, 3.1, 5.4 18
2023 arXiv
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.