---
type: source
id: naseem-2026-in-the-wild-detection-review
title: 'From curated datasets to social-media reality: In-the-wild, multimodal, and generalizable deepfake detection'
authors:
- Muhammad Tahir Naseem
- et al.
year: 2026
venue: Computer Science Review
volume: '63'
pages: '101081'
doi: 10.1016/j.cosrev.2026.101081
pmid: null
pmcid: null
arxiv_id: null
openalex_id: W7222019517
oa_url: https://doi.org/10.1016/j.cosrev.2026.101081
peer_review: peer-reviewed-journal
evidence_tier: C
design: Narrative review of deepfake detection from the perspective of in-the-wild robustness, multimodal reasoning and cross-domain generalization (no systematic protocol described in abstract)
modality:
- review
stimuli: n/a (review)
n_participants: null
n_note: Review; OpenAlex lists Wajid Zaman (Yeungnam University) as last author; full author list not confirmed
topics:
- detector-generalization
- review
summary: Review concluding that strong benchmark performance of deepfake detectors does not necessarily carry over to real-world media, because of platform compression, reposting, post-processing, modality mismatch, temporal drift and new generators; calls for deployment-oriented, continually re-benchmarked detection.
results:
- metric: Benchmark-to-wild transfer
  value: curated-benchmark performance does not necessarily translate into reliable real-world performance
  note: qualitative
  source_location: abstract
- metric: Causes of detector failure
  value: generator shift, platform-induced degradation, modality mismatch, language and speaker variation, contextual diversity
  note: ''
  source_location: abstract
- metric: Recommendations
  value: continually updated benchmarks, missing-modality robustness, uncertainty-aware detection, provenance integration
  note: ''
  source_location: abstract
moderators: []
limitations:
- Narrative review
- No quantitative synthesis in abstract
verification:
  status: verified-abstract
  confidence: high
  checked: 2026-10-19
  via: OpenAlex abstract record W7222019517 for doi:10.1016/j.cosrev.2026.101081
url: null
cited_in:
- MR-0001
updated: '2026-10-19'
---

# From curated datasets to social-media reality: In-the-wild, multimodal, and generalizable deepfake detection

> **Source record** `naseem-2026-in-the-wild-detection-review` · 2026 · Computer Science Review  
> **Evidence tier:** C (peer-reviewed-journal) · **Verification:** verified-abstract (compiler confidence: high)  
> **Bottom line:** Review concluding that strong benchmark performance of deepfake detectors does not necessarily carry over to real-world media, because of platform compression, reposting, post-processing, modality mismatch, temporal drift and new generators; calls for deployment-oriented, continually re-benchmarked detection.

## Citation {#citation}

Muhammad Tahir Naseem, et al. (2026). From curated datasets to social-media reality: In-the-wild, multimodal, and generalizable deepfake detection. *Computer Science Review*, 63, 101081. https://doi.org/10.1016/j.cosrev.2026.101081

Verification entry (identifiers, resolvers, claims with source locations): [/data/citations.json#naseem-2026-in-the-wild-detection-review](/data/citations.json#naseem-2026-in-the-wild-detection-review)

## Study design {#design}

| Field | Value |
|---|---|
| Design | Narrative review of deepfake detection from the perspective of in-the-wild robustness, multimodal reasoning and cross-domain generalization (no systematic protocol described in abstract) |
| Modality | review |
| Stimuli | n/a (review) |
| Participants (N) | not recorded |
| N detail | Review; OpenAlex lists Wajid Zaman (Yeungnam University) as last author; full author list not confirmed |
| Topics | detector-generalization, review |

## Results {#results}

| Metric | Value | Note |
|---|---|---|
| Benchmark-to-wild transfer | curated-benchmark performance does not necessarily translate into reliable real-world performance | qualitative |
| Causes of detector failure | generator shift, platform-induced degradation, modality mismatch, language and speaker variation, contextual diversity |  |
| Recommendations | continually updated benchmarks, missing-modality robustness, uncertainty-aware detection, provenance integration |  |

## Limitations {#limitations}

- Narrative review
- No quantitative synthesis in abstract

## Cited in {#cited-in}

- [MR-0001](/metareports/human-detection-accuracy.md): Human Detection of Deepfakes: Accuracy, Bias, Confidence, and Training

## Verification {#verification}

| Field | Value |
|---|---|
| Status | `verified-abstract` |
| Compiler confidence | high |
| Checked on | 2026-10-19 |
| Via | OpenAlex abstract record W7222019517 for doi:10.1016/j.cosrev.2026.101081 |

Values on this page have not all been checked against the full text. Confirm them against the primary source before high-stakes use.