ARXIV:2603.24511 · CYBERSECURITY-AI · SUBMITTED 26 MAR · 20:30 UTC · FRESHNESS STALE

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: partial proof status

Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs

Alexander Panfilov · Peter Romov · Igor Shilov · Yves-Alexandre de Montjoye · Jonas Geiping · Maksym Andriushchenko · arXiv

Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

Ship in 2-4 weeks›Score8.0Evidence partial

Opportunity summary

Pain Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

Evidence 0 refs | 0 sources | 50% coverage

Blocker Evidence partial

Open Build Read PDF Signal Canvas Track

PROBLEM

Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions. We show that an \emph{autoresearch}-style pipeline \citep{karpathy2026autoresearch} powered by Claude Code discovers novel white-box adversarial attack \textit{algorithms} that \textbf{significantly outperform all existing…

METHOD

Full abstract

LLM agents like Claude Code can not only write code but also be used for autonomous AI research and engineering \citep{rank2026posttrainbench, novikov2025alphaevolve}. We show that an \emph{autoresearch}-style pipeline \citep{karpathy2026autoresearch} powered by Claude Code discovers novel white-box adversarial attack \textit{algorithms} that \textbf{significantly outperform all existing (30+) methods} in jailbreaking and prompt injection evaluations. Starting from existing attack implementations, such as GCG~\citep{zou2023universal}, the agent iterates to produce new algorithms achieving up to 40\% attack success rate on CBRN queries against GPT-OSS-Safeguard-20B, compared to $\leq$10\% for existing algorithms (\Cref{fig:teaser}, left). The discovered algorithms generalize: attacks optimized on surrogate models transfer directly to held-out models, achieving \textbf{100\% ASR against Meta-SecAlign-70B} \citep{chen2025secalign} versus 56\% for the best baseline (\Cref{fig:teaser}, middle). Extending the findings of~\cite{carlini2025autoadvexbench}, our results are an early demonstration that incremental safety and security research can be automated using LLM agents. White-box adversarial red-teaming is particularly well-suited for this: existing methods provide strong starting points, and the optimization objective yields dense, quantitative feedback. We release all discovered attacks alongside baseline implementations and evaluation code at https://github.com/romovpa/claudini.

RESULT

ScienceToStartup currently rates this 8.0/10 on the public viability pass. We show that an \emph{autoresearch}-style pipeline \citep{karpathy2026autoresearch} powered by Claude Code discovers novel white-box adversarial attack \textit{algorithms} that \textbf{significantly outperform all existing (30+) methods}…

WHY NOW

Cybersecurity-AI moved forward this cycle; last verified April 2026. Public score 8.0/10. Implementation evidence is present through a linked repository.

Continue into Read for claims, analysis, references, and neighboring papers.

Opportunity summary

Score8.0

PainClaudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

Evidence0 refs | 0 sources | 50% coverage

Blockerno shell-level blocker reported

Analysis summary

Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: partial proof status

Competitive landscape

Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

Segment

Cybersecurity-AI

Adoption evidence

Public code linked for build inspection

Commercial read

8.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

{ "contract_version": "paper-r2", "paper_id": "b984cb6f-af71-492b-a05e-35f418dc5dbb", "arxiv_id": "2603.24511", "canonical_route": "/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "active_tab": "synced from current hash by the drawer client", "selected_artifact": "claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "endpoints": { "paper_pack": "/api/v1/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms/paper-pack", "build_passport": "/api/v1/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms/build-passport", "mcp_resource": "sciencetostartup://surfaces/paper-workspace" } }

{ "surface": "paper", "mode": "paper", "query": "Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs", "normalized_query": "2603.24511", "route": "/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "paper_ref": "claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "topic_slug": null, "benchmark_ref": null, "dataset_ref": null }

{ "@context": "https://schema.org", "@graph": [ { "@type": "WebPage", "@id": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms#webpage", "url": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "name": "Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs", "description": "Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.", "isPartOf": { "@id": "https://sciencetostartup.com/#website" } }, { "@type": "ScholarlyArticle", "@id": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms#scholarlyArticle", "headline": "Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs", "description": "Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.", "url": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms", "sameAs": "https://arxiv.org/abs/2603.24511", "identifier": { "@type": "PropertyValue", "propertyID": "arXiv", "value": "2603.24511" }, "isAccessibleForFree": true, "isPartOf": { "@id": "https://sciencetostartup.com/#website" }, "datePublished": "2026-03-25T16:50:56.000Z", "author": [ { "@type": "Person", "name": "Alexander Panfilov", "affiliation": { "@type": "Organization", "name": "Max Planck Institute for Intelligent Systems" } }, { "@type": "Person", "name": "Peter Romov", "affiliation": { "@type": "Organization", "name": "Imperial College London" } }, { "@type": "Person", "name": "Igor Shilov", "affiliation": { "@type": "Organization", "name": "Imperial College London" } }, { "@type": "Person", "name": "Yves-Alexandre de Montjoye", "affiliation": { "@type": "Organization", "name": "Imperial College London" } }, { "@type": "Person", "name": "Jonas Geiping", "affiliation": { "@type": "Organization", "name": "Max Planck Institute for Intelligent Systems" } }, { "@type": "Person", "name": "Maksym Andriushchenko", "affiliation": { "@type": "Organization", "name": "Max Planck Institute for Intelligent Systems" } } ], "codeRepository": "https://github.com/romovpa/claudini", "additionalProperty": [ { "@type": "PropertyValue", "propertyID": "viabilityScore", "value": 8 }, { "@type": "PropertyValue", "propertyID": "researchDomain", "value": "Cybersecurity-AI" }, { "@type": "PropertyValue", "propertyID": "commercialReadiness", "value": "code, repo url" } ] }, { "@type": "SoftwareSourceCode", "@id": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms#software", "name": "Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs - Source Code", "description": "Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.", "codeRepository": "https://github.com/romovpa/claudini", "url": "https://github.com/romovpa/claudini" }, { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://sciencetostartup.com" }, { "@type": "ListItem", "position": 2, "name": "Cybersecurity-AI", "item": "https://sciencetostartup.com/topics" }, { "@type": "ListItem", "position": 3, "name": "Claudini: Autoresearch Discovers State-of-the-Art Adversaria", "item": "https://sciencetostartup.com/paper/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms" } ] }, { "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "What is the startup potential of \"Claudini: Autoresearch Discovers State-of-the-Art Adversaria\"?", "acceptedAnswer": { "@type": "Answer", "text": "Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions." } }, { "@type": "Question", "name": "What products could be built from this research?", "acceptedAnswer": { "@type": "Answer", "text": "Package Claudini as a subscription-based cybersecurity tool that provides continuous testing and improvement of AI models against adversarial attacks." } }, { "@type": "Question", "name": "What are the practical use cases?", "acceptedAnswer": { "@type": "Answer", "text": "Provide cybersecurity firms with automated tools to test and improve the security of their language models against adversarial attacks." } }, { "@type": "Question", "name": "What industries could this research disrupt?", "acceptedAnswer": { "@type": "Answer", "text": "Claudini replaces traditional manually designed adversarial attacks with AI-driven automated discovery, offering faster and more effective security solutions." } } ] } ] }

Competitive landscape

Claudini autonomously discovers advanced adversarial attacks on LLMs, offering cutting-edge cybersecurity solutions.

Segment

Cybersecurity-AI

Adoption evidence

Public code linked for build inspection

Commercial read

8.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs

Claudini: Autoresearch Discovers State-of-the-Art Adversarial Attack Algorithms for LLMs

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline