Back to feed
News
Near-term (1-2 years)
January 14, 2026

New Research: 64% of 3rd-Party Applications Access Sensitive Data Without Justification

23 hours agoinfo@thehackernews.com (The Hacker News)

Summary

This research highlights a critical risk for AI systems: widespread, unjustified access to sensitive data by third-party applications, creating vulnerabilities for adversarial attacks that could compromise AI models and their outputs. With 64% of third-party apps accessing sensitive data without justification, the risk of data poisoning, model theft, or manipulation of AI-driven decision-making significantly increases, demanding stronger AI security protocols and ethical considerations.

Impact Areas

cost
risk
strategic

Sector Impact

Cybersecurity & AI Safety: This research is a direct call to action for the cybersecurity industry to develop more robust solutions for protecting AI systems from third-party vulnerabilities. The increased risk of malicious activity from the government sector further emphasizes the need for strong national security measures within the AI domain. Education & EdTech: Compromised education sites can expose sensitive student data and also corrupt training data for education AI tools. Government & Public Sector: The increased malicious activity highlights the vulnerability of government AI systems and the need for increased security investment and personnel.

Analysis Perspective
Executive Perspective

Businesses must implement AI-powered data governance and security solutions to monitor third-party data access and prevent data breaches that could contaminate training data or expose sensitive information used in AI applications. Integrating anomaly detection and predictive threat analysis into data pipelines will become critical for maintaining AI model integrity and operational efficiency.

Related Articles
News
September 22, 2022
Building safer dialogue agents  Google DeepMind
News
10 hours ago
Emversity has raised $30 million in a new round as it scales job-ready training in India.
News
July 25, 2024
AI achieves silver-medal standard solving International Mathematical Olympiad problems  Google DeepMind
Research
March 16, 2022
GopherCite: Teaching language models to support answers with verified quotes  Google DeepMind