IF-Guide

IF-Guide: Influence Function-Guided Detoxification of LLMs NeurIPS 2025 This repository contains the code for IF-Guide, the LLM detoxification technique introduced in our paper: IF-Guide: Influence Function-Guided Detoxification of LLMs Zachary Coalson , Juhan Bae, Nicholas Carlini, Sanghyun Hong...

HDI-GraphRAG-Attack

Hop-Decayed Influence: New Vulnerabilities of Structural Auxiliary Indexing in GraphRAG Pipelines with LLM Jisung Park, John Le, Heath Cooper Institute of Cybersecurity and Cryptology iC², University of Wollongong IFIP SEC 2026 jspacific0307/HDI-graphragpipeline...

CVE-2026-102551 Improper verification of the source of a communication channel in FreeRTOS-Plus-TCP Router Advertisement handling

Improper verification of the source of a communication channel in the IPv6 Router Advertisement handler in Amazon FreeRTOS-Plus-TCP before V4.4.2 might allow a remote third-party to influence an endpoint's IPv6 default gateway, on-link prefix, and address configuration via a crafted Router...

CVE-2026-102551 Improper verification of the source of a communication channel in FreeRTOS-Plus-TCP Router Advertisement handling

Improper verification of the source of a communication channel in the IPv6 Router Advertisement handler in Amazon FreeRTOS-Plus-TCP before V4.4.2 might allow a remote third-party to influence an endpoint's IPv6 default gateway, on-link prefix, and address configuration via a crafted Router...

CVE-2026-102551 Improper verification of the source of a communication channel in FreeRTOS-Plus-TCP Router Advertisement handling

Improper verification of the source of a communication channel in the IPv6 Router Advertisement handler in Amazon FreeRTOS-Plus-TCP before V4.4.2 might allow a remote third-party to influence an endpoint's IPv6 default gateway, on-link prefix, and address configuration via a crafted Router...

CVE-2026-102551

Amazon FreeRTOS-Plus-TCP is vulnerable to improper verification of the source of a communication channel in its IPv6 Router Advertisement handler prior to version V4.4.2 . A remote third-party could exploit this by sending a crafted Router Advertisement to influence an endpoint's IPv6 default gat...

CVE-2026-106160 Prototype Pollution via Model.accept and kendo.setter in Kendo UI for jQuery

In Progress® Kendo® UI for jQuery prior to 2026.3.1015, Model.accept passes server-response keys to kendo.setter without guarding against unsafe object paths containing proto, constructor, or prototype. An attacker who can supply data accepted by an affected model can modify object prototype...

Nudge the Judge: Uncovering Program Decision Hijacking in Real-World Jev Applications

System One models such as Jev change how software makes decisions by mapping open-ended semantic contexts into structured judgments that directly drive program execution. However, this paradigm introduces a new security risk, which we term Jev Program Decision Hijacking JPDH: attacker-controlled...

CVE-2026-106559

The Backstage framework is vulnerable in the @backstage/plugin-scaffolder-backend-module-confluence-to-markdown package prior to version 0.3.25 . The root cause is improper input validation within the Confluence to Markdown scaffolder module. This flaw allows an attacker to influence file write o...

EUVD-2026-94372

Backstage is an open framework for building developer portals. Prior to 0.3.25, the @backstage/plugin-scaffolder-backend-module-confluence-to-markdown package is affected by improper input validation in confluence to markdown scaffolder module. Insufficient input validation in the Confluence to...

Visual Memory Attacks Can Persist through the KV Cache

Modern language model systems operate autonomously over increasingly long contexts containing untrusted text and images. Can an adversarial input continue to steer a model even after that input is removed from its context? We show that attacks can be trained to persist through the key/value KV...

Hop-Decayed Influence: New Vulnerabilities of Structural Auxiliary Indexing in GraphRAG Pipelines with LLM

GraphRAG pipelines construct auxiliary structures during offline indexing--semantic summaries, hierarchical edges, and pre-computed scores--that determine how retrieval is prioritised at query time. Prior attacks target only instance-level components nodes, edges, triples, overlooking these...

MADBench: Benchmarking the Security of Multi-Agent Debate

Multi-agent debate MAD can improve large language model LLM reasoning by allowing multiple agents to exchange and critique their answers to the same task. However, the interactions that enable agents to correct mistakes can also spread adversarial errors and steer the agents toward an incorrect...

CVE-2026-93760

A flaw was found in Mongoid. This vulnerability allows a remote attacker, without requiring any credentials, to influence how the database evaluates queries by supplying malicious filter data. This can lead to the unintended disclosure of sensitive stored field values and may also result in reduc...

On Anthropic’s AI Misuse Report

Earlier this month, Anthropic published a long report detailing all of the Claude misuses it detected. Daniel Meissler usefully summarized the report into 117 findings. A few of the highlights: AI agents increasingly handled reconnaissance, exploitation, data theft, propaganda production,...

Astra Linux – Vulnerability in PostgresSQL 11

A vulnerability was discovered in PostgreSQL. This attack requires permission to create non-temporary objects in at least one schema, the ability to lure or wait for an administrator to create or update an affected extension in that schema, and the ability to lure or wait for a victim to use the...

Astra Linux – Vulnerability in Linux

A issue was discovered in the Linux kernel versions 2.6.39 through 5.10.16, as used in Xen. The block, net, and SCSI backends consider certain errors as ordinary bugs, which are deliberately designed to cause kernel crashes. For errors that may be influenced by guests such as memory exhaustion...

UBUNTU-CVE-2026-89554

In the Linux kernel, the following vulnerability has been resolved: mptcp: fix uninitialized localid in syncookie MPJOIN reconstruction mptcptokenjoincookieinitstate restores remotenonce, localnonce, backup, joinid, token and msk from the saved cookie entry when rebuilding the request socket for ...

CVE-2026-89554

CVE-2026-89554 affects the Linux kernel's MPTCP (Multipath TCP) implementation. When reconstructing a MP_JOIN request under SYN cookies, mptcp_token_join_cookie_init_state() fails to restore the local_id field from the saved cookie entry. Because the request-sock slab is SLAB_TYPESAFE_BY_RCU and ...

PYSEC-2026-3880 OpenHarness remote project-context commands allow persistent prompt poisoning

OpenHarness /issue and /prcomments slash commands lack remoteinvocable=False protection, allowing remote channel senders to write attacker-controlled Markdown into project context files. Admitted remote attackers can inject malicious content into .openharness/issue.md and .openharness/prcomments....