cd /news/ai-safety/deepfake-detection-is-losing-timesta… · home topics ai-safety article
[ARTICLE · art-128403] src=dev.to ↗ pub= topic=ai-safety verified=true sentiment=· neutral

Deepfake detection is losing. Timestamped proof-of-existence might be the other half of the answer

A developer writing as Kimeric argues that deepfake detection is losing ground, with some benchmarks showing failure rates around 45-50% on recent generative models, and proposes timestamped proof-of-existence systems such as OpenTimestamps as a complementary approach. The post points to a free book, time2act, which frames the problem as "detection is not enough" and connects it to the Berkeley Protocol on digital open-source investigations. Kimeric asks whether others have used proof-of-existence or timestamping schemes in forensic or content-provenance pipelines and what broke when making them verifiable offline.

by read1 min views2 publishedSep 13, 2026

I've been reading into why deepfake detectors keep losing ground (some benchmarks put failure rates around 45-50% on recent generative models), and it pushed me toward a question that's more about infrastructure than machine learning: instead of trying to prove after the fact that a piece of media is fake, what if you could cheaply prove, at the moment of capture, that it existed in a specific form at a specific time?

That's the core idea behind "proof of existence" / timestamping systems that a lot of the DFIR and blockchain-timestamping communities already work with (OpenTimestamps being the best-known example). The mechanism is simple and doesn't require trusting a central authority:

None of this proves a photo is "true" in a philosophical sense. It proves a specific bit-for-bit artifact existed at time T, which is exactly the kind of chain-of-custody primitive that forensic investigators, journalists and human-rights researchers already work with, just applied earlier in the pipeline (at capture, not at review).

I found this framed well in a free book called time2act, which digs into this from a "detection is not enough" angle and connects it to the Berkeley Protocol on digital open-source investigations. It's free, available in English/Spanish/Catalan, web/PDF/EPUB: https://www.time2act.info/read/en/

Curious whether others here have worked with proof-of-existence or timestamping schemes in forensic or content-provenance pipelines, and what broke when you tried to make them verifiable offline.

— Kimeric

(Disclosure: drafted with AI assistance, per DEV's AI content guidelines.)

── more in #ai-safety 4 stories · sorted by recency
── more on @opentimestamps 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/deepfake-detection-i…] indexed:0 read:1min 2026-09-13 ·