--- id: ai-generated-content-detection title: "AI-Generated Content Detection: Identifying Synthetic Text, Deepfake Images, and AI-Authored Media" schema_type: article category: ai language: en confidence: high last_verified: "2026-05-24" created_date: "2026-05-24" generation_method: ai_assisted ai_models: - claude-4.5-sonnet derived_from_human_seed: true conflict_of_interest: none_declared is_live_document: false data_period: static completeness: 0.85 atomic_facts: - id: af-ai-generated-content-detection-1 statement: >- A comprehensive 2025 survey of AI-generated text detection methods found that statistical detectors (GLTR, DetectGPT — analyzing token probability curvature) achieve 85-95% detection accuracy against GPT-3.5 outputs but degrade to 55-70% against GPT-4 and Claude outputs, and are easily defeated by paraphrasing (PEGASUS paraphrasing reduces detection to near-random). Watermarking techniques (Kirchenbauer et al., 2023 — biasing token sampling) provide cryptographic detection guarantees but can be removed through translation or heavy paraphrasing. source_title: AI text detection surveys (2024-2025) / Mitchell et al. (2023) — DetectGPT / Kirchenbauer et al. (2023) — Watermarking LLM outputs / Sadasivan et al. (2023) — detection-paraphrase evasion source_url: https://arxiv.org/abs/2301.11305 confidence: high - id: af-ai-generated-content-detection-2 statement: >- C2PA (Coalition for Content Provenance and Authenticity) — backed by Adobe, Microsoft, Google, OpenAI, and the BBC — published the 2.1 technical specification (2025) establishing a cryptographic chain of custody for digital media: capture devices sign content at creation, edits are recorded in a tamper-evident manifest. By 2026, flagship cameras (Leica M11-P, Sony A9 III) embed C2PA credentials at capture, and Google DeepMind's SynthID embeds imperceptible watermarks in AI-generated images, audio, and text. source_title: C2PA 2.1 Specification (2025) / Leica M11-P (2024) — first C2PA camera / Google DeepMind SynthID (2023-2025) — AI watermarking source_url: https://c2pa.org/specifications/ confidence: high primary_sources: - id: ps-ai-generated-content-detection-1 title: "DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature (and follow-up surveys)" type: academic_paper year: 2023 institution: Stanford / UC Berkeley / arXiv / ICML url: https://arxiv.org/abs/2301.11305 - id: ps-ai-generated-content-detection-2 title: "C2PA Content Credentials: Technical Specification v2.1" type: industry_standard year: 2025 institution: Coalition for Content Provenance and Authenticity (Adobe, Microsoft, Google, BBC) url: https://c2pa.org/specifications/ known_gaps: - Detecting AI-generated multimodal content (AI-generated video + AI-generated audio + AI-generated script) - Adversarial robustness — detection in the presence of deliberate evasion attacks disputed_statements: [] secondary_sources: - title: "Deepfake Generation and Detection: A Comprehensive Benchmark and Survey" type: survey_paper year: 2024 authors: - Pei, Gan - Zhang, Jiangning - et al. institution: arXiv (comprehensive review 2017-2024) url: https://arxiv.org/abs/2403.17881 - title: "A Review of Deepfake and Its Detection: From Generative Models to Detection Methods" type: survey_paper year: 2025 authors: - multiple institution: Wiley IET Journal url: https://doi.org/10.1155/int/9987535 - title: "Deepfake Detection: A Critical Review of State-of-the-Art Methods and Future Directions" type: survey_paper year: 2025 authors: - multiple institution: SN Applied Sciences (Springer) url: https://doi.org/10.1007/s42452-025-08174-9 - title: "MIT Media Lab: Detect DeepFakes — Research Project on Counteracting AI-Generated Misinformation" type: report year: 2024 authors: - MIT Media Lab institution: MIT url: https://www.media.mit.edu/projects/detect-fakes/overview/ updated: "2026-05-24" --- ## TL;DR As AI-generated text, images, audio, and video become indistinguishable from human-created content, detecting what's real becomes a critical societal challenge — from identifying AI-written student essays and fake product reviews to detecting political deepfakes. Detection is an asymmetric arms race: generation improves faster than detection, and each detection method spawns evasion techniques. ## Core Explanation AI-generated text detection approaches: (1) Statistical detectors — analyze token probability distributions. Human writing has natural variability (some words unexpected); AI-generated text has concentrated probability (model selects most likely tokens). GLTR (2019) visualizes this; DetectGPT (2023) perturbs text and measures probability curvature — AI text tends to be at local probability maxima, slight changes decrease probability; (2) Classifier-based — fine-tuned discriminators (RoBERTa, GPT-2 detector) trained on human vs. AI text pairs. Strong on specific models, poor generalization; (3) Watermarking — during generation, bias token selection toward a "green list" of tokens. Detector checks if text has disproportionately many green-list tokens — provides statistical guarantee. SynthID (Google, 2023-2025) embeds imperceptible watermarks in generated text, images, and audio; (4) Retrieval-based — store all generated text in a database, check if candidate text matches. Only works for known generators. ## Detailed Analysis Deepfake detection: (A) Image — face artifacts (inconsistent eye reflections, asymmetric earrings, blending boundaries), frequency domain analysis (GAN fingerprints), and physiological signals (heart rate from facial color changes, inconsistent with real humans). SOTA detectors (XceptionNet, EfficientNet-based) achieve 95%+ on known generators, drop to 60-70% on unseen generators; (B) Video — temporal inconsistencies across frames (flickering, unnatural blinking patterns), audio-visual asynchrony (lip movements not matching speech); (C) Audio — vocoder artifacts, unnatural prosody patterns. The arms race: every detection method spawns evasion — paraphrasing defeats statistical text detection; adversarial noise defeats image detection; re-encoding defeats watermarking. C2PA represents a proactive approach — instead of detecting fakes, establish authenticity for real content. Cameras sign images at capture with device attestation; editing software (Photoshop) adds edit claims; social media platforms display "content credentials" showing provenance chain. SynthID embeds watermarks during generation rather than post-hoc. Key open problem: the asymmetry — detection must catch 100% of fakes (one missed deepfake can cause significant harm), while generation needs only one success. Regulatory landscape: EU AI Act requires AI-generated content labeling; China mandates watermarking of deep synthesis; US AI executive orders encourage provenance standards. The consensus: no single detection method suffices — defense-in-depth combining multiple orthogonal approaches is necessary. ## Further Reading - SynthID: Google DeepMind Watermarking - C2PA/Content Authenticity Initiative (Adobe) - GLTR: Visual Forensic Tool for AI Text (Harvard/IBM)