34 matches found
How Reliable Are AI Attackers against a Fixed Vulnerable Target? A 400-Run Empirical Study of LLM Penetration Testing Consistency
Large language models LLMs can autonomously conduct multi-stage cyber attacks, but the consistency of their offensive behavior under repeated trials remains unstudied. This work presents the first large-scale empirical measurement of LLM attack consistency: 400 autonomous penetration testing runs...
The Surface You Test Is Not the Surface That Breaks
Tool-augmented LLM agents are vulnerable to prompt injection: a third party who controls part of the agent's context can plant instructions that the agent then executes as if they came from the user. Current evaluations report a single attack success rate per model on one channel, the tool output...
Refusal Before Decoding: Detecting and Exploiting Refusal Signals in Intermediate LLM Activations
In this paper, we investigate whether refusal behavior can be predicted from LLM intermediate activations before decoding using linear probes trained on residual stream activations at each transformer block. We find that refusal is linearly decodable well before the final layer, indicating that...
DarkLLM: Learning Language-Driven Adversarial Attacks with Large Language Models
While vision and multimodal foundation models underpin critical tasks from perception to complex reasoning, they remain highly vulnerable to adversarial attacks. However, traditional adversarial attacks are typically limited to single, predefined objectives, tightly coupling each attack to a...
Can You Keep a Secret? Involuntary Information Leakage in Language Model Writing
Language models are deployed in settings that require compartmentalization: system prompts should not be disclosed, chain-of-thought reasoning is hidden from users, and sensitive data passes through shared contexts. We test whether models can keep prompted information out of their writing. We giv...
VulKey: Automated Vulnerability Repair Guided by Domain-Specific Repair Patterns
The increasing prevalence of software vulnerabilities highlights the need for effective Automatic Vulnerability Repair AVR tools. While LLM-based approaches are promising, they struggle to incorporate structured security knowledge from sources like CWE and NVD. Current methods either use this...
Stop Fixating on Prompts: Reasoning Hijacking and Constraint Tightening for Red-Teaming LLM Agents
With the widespread application of LLM-based agents across various domains, their complexity has introduced new security threats. Existing red-team methods mostly rely on modifying user prompts, which lack adaptability to new data and may impact the agent's performance. To address the challenge,...
EUVD-2025-209209
Juju: Read All Controller Logs From Compromised Workload...
Incorrect Authorization
Overview Affected versions of this package are vulnerable to Incorrect Authorization debug log endpoint in the API server. An attacker can access sensitive log data belonging to any entity across any model by compromising a workload machine under the controller. Remediation A fix was pushed into...
CVE-2025-68152
CVE-2025-68152 affects Juju, an open source application orchestration engine. The issue permits a compromised workload machine under a Juju controller to read any log file for any entity in any model at any level, affecting versions 2.9 to before 2.9.56 and 3.6 to before 3.6.19. The flaw is mitig...
Not All Tokens Are Created Equal: Query-Efficient Jailbreak Fuzzing for LLMs
Large Language ModelsLLMs are widely deployed, yet are vulnerable to jailbreak prompts that elicit policy-violating outputs. Although prior studies have uncovered these risks, they typically treat all tokens as equally important during prompt mutation, overlooking the varying contributions of...
SUSE CVE-2026-1237
Vulnerable cross-model authorization in juju. If a charm's cross-model permissions are revoked or expire, a malicious user who is able to update database records can mint an invalid macaroon that is incorrectly validated by the juju controller, enabling a charm to maintain otherwise revoked or...
Juju has broken CMR authorization
Impact Cross-model Relation authorization is broken and has a potential security vulnerability. If the controller does not have the root key to verify the macaroon or if the macaroon has expired, an unvalidated and therefore untrusted macaroon is used to extract declared caveats. Facts from these...
EUVD-2026-4900
Juju has broken CMR authorization...
GHSA-J477-6VPG-6C8X Juju has broken CMR authorization
Impact Cross-model Relation authorization is broken and has a potential security vulnerability. If the controller does not have the root key to verify the macaroon or if the macaroon has expired, an unvalidated and therefore untrusted macaroon is used to extract declared caveats. Facts from these...
CVE-2026-1237
Vulnerable cross-model authorization in juju. If a charm's cross-model permissions are revoked or expire, a malicious user who is able to update database records can mint an invalid macaroon that is incorrectly validated by the juju controller, enabling a charm to maintain otherwise revoked or...
Operation on a Resource after Expiration or Release
Overview Affected versions of this package are vulnerable to Operation on a Resource after Expiration or Release in the macaroon validation for cross-model authorization. An attacker can maintain unauthorized access to resources by crafting and submitting an invalid macaroon that is incorrectly...
CVE-2026-1237
Vulnerable cross-model authorization in juju. If a charm's cross-model permissions are revoked or expire, a malicious user who is able to update database records can mint an invalid macaroon that is incorrectly validated by the juju controller, enabling a charm to maintain otherwise revoked or...
CVE-2026-1237
Vulnerable cross-model authorization in juju. If a charm's cross-model permissions are revoked or expire, a malicious user who is able to update database records can mint an invalid macaroon that is incorrectly validated by the juju controller, enabling a charm to maintain otherwise revoked or...
CVE-2026-1237
Vulnerable cross-model authorization in juju. If a charm's cross-model permissions are revoked or expire, a malicious user who is able to update database records can mint an invalid macaroon that is incorrectly validated by the juju controller, enabling a charm to maintain otherwise revoked or...