From Forgeries to Foundation Models: A Systematic Survey of Identity Document Attack and Detection

2026-07-01 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

A systematic survey, "From Forgeries to Foundation Models," analyzes the fundamental shift in identity document forgery. Generative AI tools now enable high-fidelity synthesis and manipulation, while detection methods are constrained by outdated benchmarks. The survey unifies Presentation Attacks, Digital Injection Attacks, and GenAI-driven synthesis within a single threat model. It traces detection methodologies from rule-based heuristics to foundation models. An audit of 2019-2025 public datasets reveals a "Reality Gap" between benchmarks and operational deployment. The survey identifies "Script-Dependent Generative Instability (SDGI)" in non-Latin script inpainting. It also notes that even strong models achieve APCER values above 25% on unseen synthesized IDs.

Key takeaway

For AI Security Engineers developing identity verification systems, you must urgently update your threat models and evaluation frameworks. Current detection methods are insufficient against advanced generative AI attacks. This is evidenced by the "Reality Gap" in benchmarks and APCER values exceeding 25% for strong models. Prioritize forensically grounded, privacy-preserving, and legally accountable systems to counter this evolving threat.

Key insights

Generative AI has fundamentally shifted identity document forgery, rendering current detection benchmarks inadequate against advanced attacks.

Principles

• Unified threat models are essential for diverse attack surfaces
• Public datasets exhibit a "Reality Gap" with operational threats
• Non-Latin scripts show "Script-Dependent Generative Instability"

Method

The survey systematically unifies attack types, traces detection methodologies, audits public datasets (2019-2025), analyzes large multimodal models, and performs zero-shot benchmarking on unseen synthesized IDs.

In practice

• Audit existing benchmarks for "Reality Gap" against GenAI threats
• Test detection systems with non-Latin script forgeries
• Perform zero-shot benchmarking on unseen synthetic IDs

Topics

• Identity Document Forgery
• Generative AI
• Attack Detection
• Threat Modeling
• Foundation Models
• Benchmarking

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Security Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• Friday Squid Blogging: Regulating Squid Fishing in the South Pacific — The "gap twixt physical object and sensor" makes digital identity verification inherently vulnerable to circumvention.
• Generative AI and Digital Ecosystem Resilience: A Proactive Lifecycle-Based Survey — Proactive, lifecycle-based detection of GenAI-driven inauthentic narratives is crucial for digital ecosystem resilience, shifting from reactive methods.
• From Vision to Text: A Compact Multimodal Approach for Robust, Cross-Domain Presentation Attack Detection on ID Cards — Compact multimodal PAD models need real-world data and fine-tuning for robust cross-domain performance, challenging synthetic dataset utility.
• Continuous Offensive Security: The Line We've Been Walking — AI-driven offensive security is crucial for continuous defense against autonomous attackers and new AI-specific attack surfaces.
• Adversarial Creation and Detection of AI-Generated Social Bot Content — Adversarial training with paired human and AI-generated data significantly improves social bot content detection.
• With $1 Cyberattacks on the Rise, Durable Defenses Pay Off — Generative AI accelerates both cyberattacks and defenses, necessitating a shift to foundational security rather than reactive patching.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.