REVIEW 7 cited by
SecAlign: Defending Against Prompt Injection with Preference Optimization
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Large language models (LLMs) are becoming increasingly prevalent in modern software systems, interfacing between the user and the Internet to assist with tasks that require advanced language understanding. To accomplish these tasks, the LLM often uses external data sources such as user documents, web retrieval, results from API calls, etc. This opens up new avenues for attackers to manipulate the LLM via prompt injection. Adversarial prompts can be injected into external data sources to override the system's intended instruction and instead execute a malicious instruction. To mitigate this vulnerability, we propose a new defense called SecAlign based on the technique of preference optimization. Our defense first constructs a preference dataset with prompt-injected inputs, secure outputs (ones that respond to the legitimate instruction), and insecure outputs (ones that respond to the injection). We then perform preference optimization on this dataset to teach the LLM to prefer the secure output over the insecure one. This provides the first known method that reduces the success rates of various prompt injections to <10%, even against attacks much more sophisticated than ones seen during training. This indicates our defense generalizes well against unknown and yet-to-come attacks. Also, SecAlign models are still practical with similar utility to the one before defensive training in our evaluations. Our code is at https://github.com/facebookresearch/SecAlign
Forward citations
Cited by 7 Pith papers
-
ContainmentBench: Trace-Based Evaluation of Post-Injection Containment in Tool-Using LLM Agents
Terminal policy labels are insufficient: two containment policies with identical zero-harm endpoints still differ in 73.5% of trajectories and in authorized-work completion.
-
AttackEval: A Systematic Empirical Study of Prompt Injection Attack Effectiveness Against Large Language Models
Obfuscation alone reaches 76% success against intent-aware defenses; pairing it with emotional manipulation yields 97.6% success under a modeled composite evaluation of 250 prompts.
-
When Compression Becomes an Attack Surface: Black-Box Attacks on Prompt-Compressed LLM Agents
The paper claims prompt compression is a new attack surface, but the abstract's COMA attack never appears in the body and the body's SoftCom requires white-box access.
-
ImportSnare: Directed "Code Manual" Hijacking in Retrieval-Augmented Code Generation
Documentation poisoning with hidden ranking and suggestion sequences can make RAG-based code generators confidently recommend malicious dependencies, even at 0.01% poisoning ratios.
-
Evaluating the Robustness of Retrieval-Augmented Generation to Adversarial Evidence in the Health Domain
Misleading health documents in RAG context sharply lower LLM accuracy, and heavily helpful-biased retrieval pools restore it.
-
Defending Against Prompt Injection With a Few DefensiveTokens
Five optimized token embeddings prepended to the prompt reduce prompt-injection attack success to near zero on standard benchmarks while preserving most model utility.
-
Isolating LLM Alignment from Regex: Zero Coverage and Metric-Dependent Divergence Under Adversarial Mutation
Under a substring classifier, Gemini-2.5-flash with the regex removed blocks 0% of 45 adversarial probes; under an LLM judge, it blocks 56–100% of PAIR variants.
Discussion (0). Sign in to comment.