Your Intelligent End-to-End Multi-Modal Risk Assessment AI Expert
ForensicsAI is an advanced AI-powered risk assessment agent dedicated to preventing multimedia risks through intelligent automation. As an autonomous expert system, it intelligently analyzes, reasons, and makes decisions across identity, document, and content security domains via end-to-end omni-modal understanding.
ForensicsAI operates as a sophisticated AI Agent, it proactively identifies risks, contextually adapts to new threats, learns from patterns, and provides intelligent recommendations across the entire multimedia security chain.
-
[2026-05] 🏆 GenText-Forensics Challenge @ ACM Multimedia 2026 We are organizing GenText-Forensics: Challenge on Explainable Forensics and Adversarial Generation for Text-Centric Images at ACM MM 2026 (Rio de Janeiro, 10–14 Nov 2026) — the first AI security challenge dedicated to global multilingual text-image forensics. The challenge reframes document forgery detection as a unified generative "forgery analysis" task spanning detection, spatial grounding, and natural-language explanation. Test data is released and evaluation is open on Codabench, supported by the RealText-V2 benchmark. Challenge Page →
-
[2026-04] 📄 DocShield: Evidence-Grounded Agentic Reasoning DocShield: Towards AI Document Safety via Evidence-Grounded Agentic Reasoning — the first unified framework that formulates text-centric forgery analysis as a visual-logical co-reasoning problem. A novel Cross-Cues-aware Chain of Thought (CCT) iteratively cross-validates visual anomalies with textual semantics, and a Weighted Multi-Task Reward aligns reasoning structure, spatial evidence, and authenticity prediction via GRPO. DocShield improves macro-average F1 by 41.4% over specialized frameworks and 23.4% over GPT-4o on T-IC13, with consistent gains on T-SROIE. [ arxiv ]
-
[2026-04] 🤗 DocShield Models Released (7B / 9B) Public checkpoints of the DocShield framework: DocShield-7B and DocShield-9B — vision-language models fine-tuned with the CCT reasoning mechanism and GRPO-based multi-task reward for evidence-grounded document forgery analysis. DocShield-7B → · DocShield-9B →
-
[2026-04] 📦 RealText Datasets Released (V1 / V2) RealText-V1 is a text-centric image forgery analysis dataset built to benchmark visual-logical co-reasoning over text-centric image forgeries. It pairs forged and pristine document-like text images with pixel-level manipulation masks and expert-level natural-language explanations that ground every verdict in observable visual and logical evidence. RealText-V1 is the dataset constructed and released alongside DocShield — a unified framework that formulates text-centric forgery analysis as a visual-logical co-reasoning problem via a Cross-Cues-aware Chain of Thought (CCT) and GRPO optimization with a Weighted Multi-Task Reward.
RealText-V2: A Large-Scale Multilingual Document Forgery Analysis Benchmark — a large-scale multilingual document benchmark dataset purpose-built for multilingual text image forgery analysis, pioneering in both scale and annotation depth.
- 20K+ images: a large-scale benchmark, surpassing existing document forgery analysis datasets by orders of magnitude.
- 6 languages: English, Chinese, Arabic, Thai, Malay, and Indonesian — spanning Latin, logographic, Arabic, and Thai script systems, each presenting unique forgery analysis challenges.
- 6 domains: finance, education, healthcare, live streaming, e-commerce, and natural scenes.
- Multi-granularity forgery: character-level, word-level, and semantic-level tampering.
- Multi-source samples: real-world and AIGC-synthesized forgery samples covering diverse generation pipelines.
- Rich multi-task annotations: pixel-level localization masks, tampering type labels, and expert-level natural-language explanations. RealText-V1 → · RealText-V2 →
-
[2026-02] 🏆 Challenge Organizers: DCIC 2026 Image Forgery Challenge We are organizing the inaugural "Image Forgery Analysis Challenge Based on Multi-Modal Large Models" at Digital China Innovative Competition (DCIC 2026)—Currently Underway! Join the Challenge →
-
[2025-12] LogicLens: Visual-Textual Co-Reasoning 📄
LogicLens: Visual-Textual Co-reasoning for Text-Centric Forgery Analysis [ arxiv ]
| Traditional Platform | ForensicsAI Agent |
|---|---|
| Reactive - Responds to queries | Proactive - Identifies threats autonomously |
| Isolated Analysis - Single-domain focus | Holistic Understanding - Multi-modal contextual reasoning |
| Static Rules - Fixed detection logic | Intelligent Adaptation - Learns and evolves threat patterns |
| Manual Investigation - Requires human intervention | Autonomous Reasoning - Makes intelligent recommendations |
| Point Solutions - Handles specific tasks | End-to-End Coverage - From acquisition to verdict |
ForensicsAI operates as an intelligent expert system that:
- 🧠 Reasons intelligently across text, image, audio, and video modalities
- 🎯 Makes autonomous decisions with confidence scoring and explanations
- 🔄 Adapts continuously to emerging threats and attack patterns
- 📊 Provides expert analysis with forensic-grade evidence chain management
- ⚡ Acts end-to-end across the entire multimedia security lifecycle
- 🧠 Intelligent multi-modal identity reasoning and authentication
- 🎯 Autonomous fake account detection with behavioral analysis
- 📈 Adaptive liveness detection against evolving spoofing attacks
- 🔐 Biometric feature protection with forensic-grade analysis
- 🔍 Forensic-grade image forgery analysis with intelligent tamper reasoning
- 🎨 Visual-textual co-reasoning for document authenticity assessment
- 📊 Intelligent metadata analysis and source tracing
- 🔎 Adaptive detection against sophisticated forgery techniques
- 🤖 Advanced AIGC generation intelligence - understands synthesis patterns
- 🎬 Deepfake detection through behavioral and artifact analysis
- 🗣️ Intelligent content risk assessment with context understanding
- 🚨 Real-time threat detection with automatic flagging
- 🎯 Intelligent decision engine - synthesizes insights across all modalities
- 🧠 Expert reasoning system - provides forensic-grade analysis and justification
- 📈 Confidence scoring - quantifies threat level with explainability
- 🔄 Continuous intelligence - learns from patterns and evolves defenses
- ⚡ Zero-trust verification - validates across entire multimedia chain
- ✅ Autonomous Intelligence - Makes decisions beyond traditional rule-based systems
- ✅ Multi-Modal Co-Reasoning - Understands relationships across all media types
- ✅ Adaptive Threat Detection - Evolves against new attack patterns in real-time
- ✅ Explainable AI - Provides forensic-grade justification for every verdict
- ✅ End-to-End Verification - Validates entire content lifecycle
- ✅ Expert-Grade Analysis - Mimics and augments human forensic expert capabilities
pip install forensics-aifrom forensics_ai import IdentitySecurityAgent
# Initialize the intelligent identity expert
id_agent = IdentitySecurityAgent(mode="expert")
# Autonomously analyze identity claims
result = id_agent.verify({
"identity_images": ["selfie.jpg", "id_card.jpg"],
"metadata": {"timestamp": "2026-03-04", "device_id": "..."},
"behavioral_signals": {...}
})
# Agent provides intelligent reasoning
print(f"Risk Assessment: {result['risk_score']:.2%}")
print(f"Confidence: {result['confidence']:.2%}")
print(f"Expert Analysis: {result['reasoning']}")
print(f"Recommended Action: {result['recommendation']}")from forensics_ai import DocumentForensicsAgent
# Initialize visual-textual reasoning agent
doc_agent = DocumentForensicsAgent(model="logicLens")
# Intelligent analysis combining visual & textual cues
analysis = doc_agent.investigate({
"document": "contract.pdf",
"context": "legal_verification",
"threat_model": "advanced"
})
# Receives expert-level forensic report
print(f"Forgery Detection: {analysis['forgery_probability']:.2%}")
print(f"Tamper Regions: {analysis['tampered_areas']}")
print(f"Forensic Evidence: {analysis['evidence_chain']}")
print(f"Expert Verdict: {analysis['verdict']}")from forensics_ai import ContentSecurityAgent
# Initialize autonomous content expert
content_agent = ContentSecurityAgent()
# End-to-end multi-modal threat assessment
threats = content_agent.scan_multimodal({
"image": "social_media_post.jpg",
"video": "video_clip.mp4",
"audio": "audio_track.wav",
"text": "accompanying_caption.txt"
})
# Intelligent synthesis across modalities
print(f"AIGC Risk: {threats['generation_risk']:.2%}")
print(f"Content Compliance: {threats['compliance_score']:.2%}")
print(f"Multi-Modal Consistency: {threats['consistency_analysis']}")
print(f"Recommended Actions: {threats['action_items']}")from forensics_ai import RiskOrchestrationAgent
# Initialize the comprehensive risk expert
risk_agent = RiskOrchestrationAgent(expert_mode=True)
# Autonomous investigation across entire multimedia chain
investigation = risk_agent.orchestrate({
"source": "user_submission",
"content_type": "multimodal",
"context": {"user_history": {...}, "market_context": {...}},
"investigation_depth": "comprehensive"
})
# Receives integrated expert assessment
print(f"Overall Risk Score: {investigation['risk_verdict']:.2%}")
print(f"Threat Classification: {investigation['threat_type']}")
print(f"Confidence Level: {investigation['confidence']:.2%}")
print(f"Forensic Chain: {investigation['evidence_report']}")
print(f"Recommended Policy: {investigation['policy_recommendation']}")ForensicsAI agents operate autonomously in high-stakes scenarios where intelligent decision-making is critical:
-
Financial Risk Control: Autonomous KYC/KYB verification with expert-grade confidence scoring
- Makes intelligent decisions on account approvals with forensic grading
- Continuously adapts to new fraud patterns
-
Social Platform Governance: Intelligent content moderation agents
- Autonomous decision-making on harmful content
- Context-aware understanding of intent and impact
- Explains decisions to content creators and reviewers
-
Media & Copyright Protection: Expert forensic investigation agents
- Autonomously traces content origins across platforms
- Intelligent deepfake prosecution support
- Provides admissible forensic evidence chains
-
Government & Compliance: Authoritative document verification agents
- Autonomous certificate and credential verification
- Expert-level analysis for official proceedings
- Forensic report generation for legal purposes
-
Law Enforcement: Forensic intelligence agents
- Autonomous audio-visual evidence authentication
- Evidence chain management with expert certifications
- Supports investigative workflows with explainable findings
-
AIGC Compliance & Safety: Content generation monitoring agents
- Real-time autonomous detection of synthetic media generation
- Intelligent classification of AI-generated vs. authentic content
- Provides transparency and compliance labeling
| Security Domain | Identity Intelligence | Document Forensics | Content Analysis | Risk Orchestration |
|---|---|---|---|---|
| Reasoning Type | Behavioral & Biometric | Visual-Textual Co-reasoning | Multi-modal Pattern Recognition | Integrated Holistic Assessment |
| Decision Making | Autonomous Verification | Intelligent Verdict | Threat Classification | Policy Recommendation |
| Adaptability | Learns spoofing patterns | Learns forgery techniques | Adapts to new generators | Evolves threat models |
| Explainability | Confidence scoring | Forensic evidence chain | Artifact evidence | Integrated reasoning trace |
Coming Soon - Repository architecture diagram will be added here
- 📖 Agent Configuration Guide
- 🤗 HuggingFace Agent Hub
- 🧠 Reasoning Engine Documentation
- 💬 Community Discussions
- 🐛 Issue Tracker & Agent Feedback
This project is licensed under the Apache License 2.0. See LICENSE file for details.
We welcome contributions that enhance our AI agents' intelligence and capabilities! Areas for contribution:
- Agent Improvements: Enhance reasoning, decision-making, and adaptability
- Model Development: Contribute improved foundation models for agents
- Explainability: Help make agent reasoning more transparent and interpretable
- Forensic Research: Contribute domain expertise to improve agent accuracy
- Use Case Integration: Build agent integrations for new domains
Contribution Process:
- Fork this repository
- Create your feature branch (
git checkout -b feature/EnhancedAgentReasoning) - Commit your improvements (
git commit -m 'Enhance agent intelligence for X domain') - Push to the branch (
git push origin feature/EnhancedAgentReasoning) - Open a Pull Request describing your agent improvements
- 📧 Email: uestczfw@gamil.com / riczuefwsct@alu.uestc.edu.cn
- 🌐 Website: forensics-ai.git.io
- 💼 Enterprise & Research Partnerships: uestczfw@gamil.com / riczuefwsct@alu.uestc.edu.cn
We thank our research community, contributors, and enterprise partners for advancing the frontier of intelligent multimedia security through agentic AI.
ForensicsAI is not just a tool—it's an autonomous intelligence partner for multimedia risk prevention. As AI-powered agents become integral to security operations, we're committed to building agents that reason like experts, act autonomously, and continuously adapt to emerging threats in the multimedia landscape.
ForensicsAI - Intelligent AI Agents Preventing Multimedia Risks
