How Jailbreak Attacks Inform Safety Alignment: A Defender-Centric, Shapley-Based Evaluation of Jailbreak Contributions

Jailbreak attacks on large language models are usually evaluated by attacker-centric metrics such as attack success rate ASR, yet an attack that breaks a model is not necessarily useful for improving its safety. We propose a defender-centric view of jailbreak evaluation, where attacks are evaluat...

Duplicate Advisory: Untrusted Query Object Evaluation in RPC API

Duplicate Advisory This advisory has been withdrawn because it is a duplicate of GHSA-64f8-pjgr-9wmr. This link is maintained to preserve external references. Original Description SurrealDB before 1.5.5 and 2.0.0-beta before 2.0.0-beta.3 accepts an arbitrary object in the signin and signup...

GHSA-XW2C-HRR9-4XWR Duplicate Advisory: Untrusted Query Object Evaluation in RPC API

Duplicate Advisory This advisory has been withdrawn because it is a duplicate of GHSA-64f8-pjgr-9wmr. This link is maintained to preserve external references. Original Description SurrealDB before 1.5.5 and 2.0.0-beta before 2.0.0-beta.3 accepts an arbitrary object in the signin and signup...

Adaptive Incident Prioritization for Security Operations at Scale

Large security operations centers SOCs often face hundreds of active incidents per day, creating substantial cognitive and operational demands for analysts. Analysts must quickly decide which incidents deserve attention within long, constantly changing queues, yet incidents are commonly ordered b...

How Do You Choose Your AI Component? an Interview Study of Secure AI Integration in Practice

The increasing adoption of Large Language Models LLMs as AI components in modern software systems introduces distinct security risks to the software supply chain. While many considerations and safety mechanisms are in place for components of the traditional software supply chain, the recent rapid...

Skipper: Incomplete fix for CVE-2026-50197: an oversized body can bypass OPA deny-on-presence Rego policies

Summary A wrong policy can be an open door. You have to check input.attributes.request.http.truncatedbody in your policy. Description Incomplete fix for CVE-2026-50197: an oversized declared-Content-Length body still hands OPA an empty parsedbody, so deny-on-presence Rego policies fail OPEN while...

CVE-2026-55254 NCalc: Denial of Service via Unbounded and Non-Terminating Factorial Evaluation

NCalc is a fast, lightweight expression evaluator for .NET. Prior to 6.1.1, the factorial operator implementation in src/NCalc.Core/Helpers/MathHelper.cs permits specially crafted expressions with extremely large factorial operands, causing excessive CPU consumption or a non-terminating loop due ...

AI Watermark Evidence Fails Forensic Readiness: An Empirical Evaluation

Governments are increasingly mandating that LLM-generated content carry watermarks. The EU AI Act calls for markings that are "sufficiently reliable and robust." California's SB 942 requires disclosure that is "permanent or extraordinarily difficult to remove." Both mandates rest on an untested...

Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents

Security-agent evaluations commonly measure peak offensive capability under generous inference budgets, emphasizing vulnerability discovery, exploit development, penetration testing, and CTF completion. Such measurements are useful but incomplete: in operational security, every reasoning step, to...

Context Contamination in LLM Analysis of Network Security Logs: Poison with Passive Prompt Injection and Mitigation Evaluation

Large Language Models are increasingly deployed in Security Operations Centers for log analysis tasks including summarization, alert triage, and threat investigation. These systems ingest logs from external-facing services and process network logs as natural language contexts to generate security...

Agent Skill Security: Threat Models, Attacks, Defenses, and Evaluation

Reusable skills are becoming a fundamental building block of Large Language Model LLM agents, enabling capabilities to be packaged, shared, and reused across diverse applications. However, existing security research primarily focuses on prompt injection and runtime execution, leaving security ris...

Malicious code in @vitets/vite-ts (npm)

--- -= Per source details. Do not edit below this line.=- Source: amazon-inspector 8fe093d0d0fa83ab20aa57e9d9c8500e03a25ead578ff351fdc3609118cf5ecf Package is published as @vitets/vite-ts and copies the legitimate Vite project's author Evan You, README, homepage vitejs.dev, and repository...

Evaluating Frontier AI Agents As Autonomous Clinical Security Auditors

Clinical AI models can expose patients to harm when adversarial vulnerabilities go undetected, yet formal security auditing requires statistical expertise, specialized tools, and significant time. We present an open evaluation task, built on METR Task Standard v0.3.0, that tests whether frontier ...

Malicious code in font-huge (npm)

--- -= Per source details. Do not edit below this line.=- Source: amazon-inspector a529537a0875d6ab8e2c942ace1db227c7131ff06b7eb4e86b12cd4193da4879 font-huge advertises itself as a font/icon collection, but its exported getPlugin function fetches https://svganchordev.net/icons/107 and passes the...

GHSA-9JPV-C7P4-997X DIRAC is vulnerable to RCE in RequestManager due to eval on untrusted input

Summary An remote code execution vulnerability exists in RequestManager due to the use of eval on untrusted input that allows any authenticated user to run code/commands on the DIRAC server as the system user running the DIRAC services. Details The exportgetRequestCountersWeb function is callable...

PYSEC-2026-2756 Open WebUI: Mass Assignment via FeedbackForm extra=allow Allows Feedback User ID Spoofing and Evaluation Data Manipulation

Mass Assignment in Feedback Creation Allows User ID Spoofing and Evaluation Data Manipulation Summary The POST /api/v1/evaluations/feedback endpoint in Open WebUI v0.9.2 is vulnerable to mass assignment via FeedbackForm, which uses modelconfig = ConfigDictextra='allow'. Due to an insecure...

Open WebUI: Mass Assignment via FeedbackForm extra=allow Allows Feedback User ID Spoofing and Evaluation Data Manipulation

Mass Assignment in Feedback Creation Allows User ID Spoofing and Evaluation Data Manipulation SummaryThe POST /api/v1/evaluations/feedback endpoint in Open WebUI v0.9.2 is vulnerable to mass assignment via FeedbackForm, which uses modelconfig = ConfigDictextra='allow'. Due to an insecure dictiona...

PYSEC-2026-3077 Superduper: Remote code execution via unsafe eval in superduper query parsing

The superduper project thru v0.10.0 contains a critical remote code execution vulnerability in its query parsing component. The parseoppart function in query.py uses the unsafe eval function to dynamically evaluate user-supplied query operands without proper sanitization or restriction. Although...

PYSEC-2026-2477 flash-attention contains an insecure deserialization vulnerability in its checkpoint loading mechanism

The flash-attention training framework thru commit e724e2588cbe754beb97cf7c011b5e7e34119e62 2025-13-04 contains an insecure deserialization vulnerability CWE-502 in its checkpoint loading mechanism. The loadcheckpoint function in checkpoint.py and the checkpoint loading code in eval.py use...

PYSEC-2026-3402 verl's math_equal() Vulnerable to Arbitrary Code Execution via Unsafe eval()

A vulnerability was identified in ByteDance verl up to 0.7.1. Affected is the function mathequal of the file primemath/grader.py. The manipulation leads to a sandbox issue. It is possible to initiate the attack remotely. The complexity of an attack is rather high. The exploitability is told to be...