Test finds Gemini Decision 1 fails to detect newer AI voices
Start here
An independent test found that Google's Gemini Decision 1, a multimodal decision model released on November 11, 2026, can distinguish AI-generated voices from human audiobook clips with an AUC of 0.67 for generators from 2017–2024, but only 0.52 — effectively chance — for generators released in 2025–2026. The tester scored 30 clips from each of 34 generators in the MLAAD dataset against 200 LibriTTS human clips, re-encoding and normalizing the audio to strip metadata. The model also leaned heavily toward labeling audio as human-made, giving only 20 of 1,020 AI clips a 50% or higher chance of being AI-generated. The result suggests the detector's accuracy degrades sharply on newer voice generators, and that its default judgments skew toward human.
Generated from reports · updated 23 h ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AITest finds Gemini Decision 1 cannot separate AI voices made after 2024
An independent test found that Google's Gemini Decision 1, a multimodal decision model released on November 11, 2026, separates AI-generated voices from human audiobook clips with an AUC of 0.67 for generators from 2017-2024, but only 0.52, effectively chance, for generators released in 2025-2026. The author scored 30 clips from each of 34 generators in the MLAAD dataset against 200 LibriTTS human clips, re-encoding and normalizing audio to strip metadata. The model also showed a strong bias toward labeling audio as human-made, giving only 20 of 1,020 AI clips a 50% or higher chance of being AI-generated.
Heat for this event
- Current, comparable range
- 5
- Peak, comparable range
- 10November 17 10:00
- Change in 24 hours
- –
The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.