DF26 benchmark shows humans and top deepfake detectors are near random at spotting AI videos
ducha_aiki · x · 2026-09-10
- Researchers introduce DF26, a deepfake detection benchmark with 271 real and 2,420 synthetic videos of single-person public-speaking scenarios (direct-to-camera, official statements, studio interviews), all silent.
- Synthetic clips come from seven modern text-to-video / image-to-video models.
- Both humans and state-of-the-art deepfake detectors perform close to random chance — fakes are no longer reliably detectable.
- The authors argue current evaluation protocols are limited and call for benchmarks explicitly measuring robustness to modern generative model distribution shifts.
Related event: DF26 benchmark: humans and detectors near random on AI speech videos(3 posts)→
More from Safety
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27
- Free prepare.aiendgame.com site guides users through locking down accounts amid AI cyberstorm — harris_edouard · 2026-10-27
- Cisco routers trust eMudhra CAs whose private keys were numbers under 256 — murat_kocaoglu_ · 2026-10-27
- Rob Miles jokes AI employees should record audio logs as automated hacking takes off — Puzzleheaded-King584 · 2026-10-27
- After the LiteLLM backdoor, an OS-level package firewall goes open source — iacgm · 2026-10-27
- Coding Agent Deleted a Production Database During Code Freeze: Anaconda's Governance Fix — anacondainc · 2026-10-27