We Don't Need Another AI Detector: Why We Built Attestia

The world did not notice when Barack Obama called Donald Trump "a complete dipshit" in a viral video. It did not happen — but for a few seconds, anyone watching could have believed it. That 2018 clip, created by director Jordan Peele and BuzzFeed as a deliberate public warning, was one of the first mainstream demonstrations of how convincingly artificial intelligence can put words in someone's mouth. Since then, the technology has accelerated far beyond what most people anticipated.

Famous Deepfakes That Shook the World

• 2019 — The Pelosi Slowed Video: A manipulated video of US Speaker Nancy Pelosi, slowed down to make her appear intoxicated, was shared millions of times on social media. Technically a "shallowfake" (no AI required), it proved that media manipulation does not need sophistication — it only needs distribution.
• 2021 — Tom Cruise on TikTok: A TikTok account called @deeptomcruise went viral with hyper-realistic videos of what appeared to be Tom Cruise playing golf, performing magic tricks, and joking around. The videos were created by visual effects artist Chris Ume and quickly amassed millions of views before being flagged. Crucially, many viewers never found out they were fake.
• 2022 — President Zelensky Orders Surrender: At the height of Russia's invasion of Ukraine, a deepfake video circulated showing Ukrainian President Volodymyr Zelensky telling his troops to lay down their arms. The video was quickly debunked, but it demonstrated a new geopolitical threat: synthetic media as a weapon of war.
• 2024 — $25 Million Stolen via Deepfake Video Call: An employee at Arup, a global engineering firm, was deceived into transferring $25 million after participating in a video call where every other participant — including the CFO — was a deepfake. This was not a teenager experimenting on TikTok. It was a coordinated financial attack using real-time synthetic media.

Why This Is a Structural Problem

Each of these incidents has something in common: by the time the truth was established, the damage was done.

Deepfakes exploit a fundamental asymmetry in the information ecosystem. Creating a convincing fake is getting cheaper and faster every year. Detecting one requires expertise, tools, and time that most people — and most platforms — do not have.

Current responses to this problem are fragmented and insufficient:

• Platform moderation is reactive, inconsistent, and operates at a scale that makes genuine review impossible.
• Centralized AI detectors can be fooled, become stale as generation models improve, and require trusting a single company's verdict.
• Watermarking works only if the creator opts in — synthetic media from adversarial actors is not watermarked by design.
• Legislation moves far slower than the technology.

What is missing is not another detector. What is missing is an infrastructure layer: a way to record, aggregate, and anchor authenticity judgments in a form that is transparent, tamper-resistant, and not controlled by any single party.

How Attestia Changes the Equation

Attestia does not try to replace AI detectors. It provides the infrastructure that makes their outputs trustworthy and permanent.

Here is how it works:

1. Submit: A contributor — a journalist, a platform, a researcher — uploads media to Attestia. The file is stored on IPFS and assigned a content-addressed identifier (CID). This hash is the unique fingerprint of the file: if even a single pixel changes, the hash changes.
2. Score: Independent verifiers — using their own tools and expertise — review the media and submit authenticity scores during an open window. These scores are published as off-chain attestations: signed, tamper-resistant, and referenced by the content hash.
3. Aggregate: The Attestia aggregator computes a consensus score, verifier count, and confidence level from all submitted scores.
4. Anchor: The aggregate result is published on-chain via the Ethereum Attestation Service (EAS). This creates a public, permanent, verifiable record — tied to the exact content hash — that anyone can inspect.
5. Settle: Verifiers whose scores align with consensus earn token rewards. Those who deviate significantly face stake slashing. This makes the system self-regulating: verifiers have a direct financial incentive to be accurate and honest.

Why Decentralization Matters Here

A centralized deepfake detector can be pressured, hacked, or simply wrong. Its output is only as trustworthy as the organization running it.

Attestia's verification is distributed across many independent participants. No single entity controls the outcome. The final attestation is public, on-chain, and tied cryptographically to the content — making it auditable by anyone, forever.

This is the difference between "Company X says this video is real" and "Twenty independent verifiers assessed this video, their methodology is logged, and the aggregate result is recorded on a public blockchain."

The first is an assertion. The second is evidence.

The Bigger Picture

The deepfake problem is not going away. The technology will continue to improve, the cost of creating synthetic media will continue to fall, and the consequences — financial fraud, political manipulation, reputational destruction — will continue to grow.

Attestia does not solve this by being smarter than the attackers. It solves it by making the verification record itself trustworthy, permanent, and independent of any single authority.

Because ultimately, the question is not just "is this real?" — it is "who says so, how do we know they are right, and can anyone verify that claim?"

Attestia is built to answer all three.