Characterizing and Defending Utility-Hijacking in Instruction-Tuning Data Selection

Instruction-data selectors choose which instruction-following records are used for fine-tuning, improving alignment quality, raising training efficiency, and filtering out harmful records. A selector's choices thus directly affect the fine-tuned model's safety, yet despite their widespread use, t...

Towards Efficient Threat Intelligence Instruction Tuning Via Quality-Aware and Optimal Transport-Guided Data Selection

High-quality data is essential for improving instruction tuning in threat intelligence analysis. However, existing methods for generating cybersecurity instruction data lack reliable quality assessment mechanisms, and gradient influence-based data selection methods primarily focus on single-sampl...

ROS-20261002-73-0063

The vulnerability of the data selection evaluator in PostgreSQL database management systems is related to incorrect session duration. Exploiting this vulnerability can allow an attacker to disclose sensitive information...

ROS-20261002-73-0064

The vulnerability of the data selection evaluator in PostgreSQL database management systems is related to incorrect session duration. Exploiting this vulnerability can allow an attacker to disclose sensitive information...

ROS-20261002-73-0062

The vulnerability of the data selection evaluator in PostgreSQL database management systems is related to incorrect session duration. Exploiting this vulnerability can allow an attacker to disclose sensitive information...

ROS-20261002-73-0061

The vulnerability of the data selection evaluator in PostgreSQL database management systems is related to incorrect session duration. Exploiting this vulnerability can allow an attacker to disclose sensitive information...

ROS-20261002-73-0065

The vulnerability of the data selection evaluator in PostgreSQL database management systems is related to incorrect session duration. Exploiting this vulnerability can allow an attacker to disclose sensitive information...

High-Quality Data Do Not Mean Safe! Poisoning LLMs after Data Selection

Safety-aligned Large Language Models remain vulnerable to fine-tuning on small sets of harmful or benign-looking samples. However, prior studies typically assume that poisoned samples directly enter downstream fine-tuning, overlooking quality-based selection in practical training pipelines. To fi...

PostgreSQL ctid data type selector Disclosure

The data type selector vulnerability in the ctid evaluator of a PostgreSQL database management system is related to an incorrect session expiration time. Exploitation of this vulnerability may allow a remote attacker to disclose protected information...

EUVD-2021-11539

The G Auto-Hyperlink WordPress plugin through 1.0.1 does not sanitise or escape an 'id' GET parameter before using it in a SQL statement, to select data to be displayed in the admin dashboard, leading to an authenticated SQL injection...

Taffy has an unspecified vulnerability

Taffy is a data selection engine that supports insertion, updating and statistics. A security vulnerability exists in taffy 2.6.2 and earlier versions. An attacker can exploit the vulnerability to access arbitrary data entries in the DB with the help of user input with redundant attributes...