SynGhost

Syntactic-Ghost SynGhost SynGhost: 構文転移による不可視かつ汎用的なタスク非依存バックドア攻撃 貢献と特徴 SynGhost には以下の貢献があります 既存のタスク非依存バックドアのリスクを軽減するため、我々は $\mathttmaxEntropy$ を提案します。これはエントロピーに基づくポイズニングフィルタであり、ポイズンされたサンプルを正確に検出します。 PLM の脆弱性をさらに明らかにするため、我々は $\mathttSynGhost$...

rats-re

ローカルLLMによるRats!ソースコード再構築 このリポジトリは、Sean O'Connorによる Rats!1994年のオリジナルWindows版であるRATS.EXEのソースコードを再構築する進行中のプロジェクトです。wibo上でMicrosoft Visual C++ 4.1を使ってWin32実行ファイルをビルドし、DREAMMでテストできます。...

PT-2026-107230

Name of the Vulnerable Software and Affected Versions SonicWall SMA 1000 Series affected versions not specified Description A pre-authentication server-side request forgery SSRF flaw exists in the SMA1000 Appliance WorkPlace interface. This issue is caused by an unintended alternate access path...

Secure Speculative Decoding for Large Language Models

Speculative decoding accelerates inference for a large language model LLM, referred to as the target model, by first using a smaller model, referred to as the draft model, to generate candidate tokens and then verifying them with the target model for acceptance or rejection. Prior studies primari...

The Model Plants the Trigger: Answer-Side Backdoor Attacks in Multi-Turn Large Language Models

Safety alignment in Large Language Models LLMs remains vulnerable to backdoor attacks. Existing LLM backdoors are almost all input-centric: activation depends on explicit trigger patterns in the user input, so modern guardrails are built to sanitize the input space. We challenge this assumption...

gradient-untangler

gradient-untangler A local research harness that searches for the exact tokens that make an open-weight language model start its answer the way you specify. Open-weight models ship as ordinary files: config.json, a tokenizer, and one or more .safetensors or .bin shards. A Hugging Face id is only ...

PYSEC-2026-4192

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

PYSEC-2026-4192

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

CVE-2026-105745

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

CVE-2026-105746 Docling: KServe v2 OCR engine does not enforce enable_remote_services

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.83.0 until 2.131.0, the KServeV2OcrModel class defined in docling/models/stages/ocr/kservev2ocrmodel.py sends page images to its configured endpoint without checki...

CVE-2026-105745 Docling: Plugin entry points are imported before the allow_external_plugins check

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

CVE-2026-105745 Docling: Plugin entry points are imported before the allow_external_plugins check

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

CVE-2026-105745

The Docling library (versions 2.27.0 through 2.131.0 ) is vulnerable to arbitrary code execution due to an incorrect behavior order in the docling/models/factories/base_factory.py component. The software calls load_setuptools_entrypoints() before validating the allow_external_plugins setting, cau...

CVE-2026-105745 Docling: Plugin entry points are imported before the allow_external_plugins check

Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. From 2.27.0 until 2.131.0, Docling plugin factories in docling/models/factories/basefactory.py call loadsetuptoolsentrypoints before applying the allowexternalplugins...

Adaptive_Greedy_Local_Search

Adaptive Greedy Local Search AGLS 意味保存型プロンプトハイジャック：自動プロンプト最適化に対するブラックボックス敵対的攻撃（ICME 2026） 概要：...

SAFESHIELD: A Decision-Organization Framework for Deployment-Time Safety of Small Language Models

Deployment-time safety of language models is commonly implemented through runtime guardrails such as input moderation, routing, retrieval verification, and output filtering. Existing deployment frameworks provide increasingly capable mechanisms for these functions, but offer limited guidance on h...

PT-2026-106248

Name of the Vulnerable Software and Affected Versions Docling versions 2.27.0 through 2.130.0 Description Plugin factories in docling/models/factories/base factory.py call the load setuptools entrypoints function before the allow external plugins setting is applied. This sequence allows every...

An Evaluation of the Semantic Understanding Capabilities of Large Language Models for Web Attack Payloads

Computer vision services delivered through Web interfaces and APIs process textual requests for image-resource acquisition, inference-task configuration, and result management, making Web attack-payload analysis relevant to their deployment security. Large language models LLMs can identify payloa...

PT-2026-106280

Name of the Vulnerable Software and Affected Versions Joplin Server versions prior to 3.7.2 Description In the packages/server/src/models/TokenModel.ts file, CSRF, account-confirmation, email-change, and password-reset tokens are stored without a designated purpose. Additionally, the...

Backdooring Sparse Autoencoders

Sparse autoencoders SAEs are increasingly used not only to interpret language models but also to intervene on their internal representations. We show that this creates a supply-chain attack surface: a maliciously modified SAE can induce attacker-chosen behavior when inserted into the forward pass...