CINDER · June 2026
News
Synthesia partners with Cinder to power the world's safest AI video platform
Synthesia extended its trust & safety stack with Cinder so enforcement keeps pace with how fast the product and the threat landscape are moving.
7 MIN READ
the update
A different view from the start
Today Synthesia is announcing its extending its trust and safety stack by partnering with Cinder, mission-critical infrastructure protecting the world’s most innovative companies from digital abuse and manipulation.
The partnership deepens a commitment Synthesia has held since the company was founded: that enterprise AI video has to be built around trust, because the customers who depend on it cannot afford for it to be built any other way.
Cinder PoC · model evaluation
Measured against Synthesia's own enforcement history.
- 97%Precision
Accuracy of positive predictions
- 967%F1 Score
Harmonic mean of precision & recall
- 96%Recall
Coverage of actual positives
The standard
How Synthesia pioneered moderation at the point of creation
When Synthesia started building AI video almost a decade ago, the standard model for online safety was detection after the fact. Online platforms hosted content, waited for it to be flagged or reported, and then decided what to take down. Synthesia took a different view from the start. If a video was going to be generated by AI, the right moment to make a moderation decision was before the model produced a single frame.
That position was unusual at the time, and in this category it still is. From the beginning, every script submitted to Synthesia was assessed against its policies in line with the generation request itself. Synthesia’s trust and safety team became specialists in a category that barely existed, designing the rules for kinds of content the rest of the industry had not yet started worrying about.
This approach matured into a system that includes AI classifiers trained on Synthesia’s own enforcement history, and a trust and safety team providing 24/7 expert human review. Each layer catches what the others might miss. The technology delivers consistency at scale; the people handle nuance, context, and the cases that genuinely require judgement.
Building this took sustained investment from Synthesia. Since 2017, the team developed its moderation tooling entirely in-house, alongside the generative models themselves. Synthesia’s classifiers have been iteratively refined through enforcement outcomes, appeals data, internal and external red-teaming exercises, and the practical feedback of customers operating in regulated industries. Every cycle of refinement reduces friction for legitimate users and tightens detection around harmful content.
This is what turns safety from a constraint into a foundation for enterprise trust. Companies in financial services, healthcare, defence, the public sector, and others from the Fortune 100 use Synthesia because they know the platform has done the work to prevent misuse before it happens. Trust is the reason Synthesia’s customers can move quickly.
The work has earned external validation too. Synthesia passed audits and holds certifications for global standards such as ISO 27001, ISO 42001, and ISO 27701, confirming that the controls behind Synthesia’s product meet rigorous global standards.
The approach
What changes with Cinder
- 11.5M+
- Pieces of content reviewed by automated moderation in 2025, up roughly 77% year over year.
- 2.7X
- More automated removals than the prior year, from 314,403 to 841,957.
- 44%
- Less content reaching human review even as volume grew, freeing reviewers for appeals and high-stakes calls.
Synthesia’s trust and safety team has been doing this work for years and knows its policies, edge cases, and customer base better than anyone. What it needs now is infrastructure that can keep pace with how quickly the product and the threat landscape are moving, which is why Synthesia is combining its know-how with Cinder's technological capabilities.
Cinder gives Synthesia an in-house agent that acts as a second pass on every model decision, gathering context, reconciling signals, and only escalating to a human reviewer when there is genuinely a judgement call to make. Cinder’s classifiers evaluate every script in line, before a frame is rendered. Every action a reviewer takes becomes a signal that retrains the system, so the model sharpens week after week.
Equally important, Cinder is built for product velocity. When Synthesia launches new avatar types, new languages, or new product surfaces, the trust and safety team can move at the pace of the roadmap.
Each year, Synthesia publishes Futuresafe, its annual responsible creation report, so customers, regulators, and the public can see how the system performs in detail. The 2025 edition documents the scale at which Synthesia existing systems already operate, and gives a clear picture of the foundation Cinder is now extending.
In 2025, Synthesia’s automated moderation tooling reviewed more than 11.5 million pieces of content and removed 841,957 of them where they violated Synthesia’s Acceptable Use Policy or Content Moderation Policies. Manual review handled a further 382,792 items and removed 70,272. Across all customer segments, 787 accounts were terminated for repeated violations.
Automated moderation volume grew by roughly 77% between 2024 and 2025, from 6.56 million items to 11.58 million, reflecting both platform growth and a deliberate shift towards catching issues earlier in the workflow. Automated removals more than doubled in the same period, from 314,403 to 841,957. At the same time, the volume of content reaching human review fell from 792,586 to 441,086, and manual removals fell from 138,692 to 70,272. This is the pattern Synthesia has been working towards for years. AI handles the obvious cases at scale, while human reviewers concentrate on appeals, ambiguous cases, and the high-stakes decisions where their judgement matters most.
Trust in Synthesia’s appeals process shows up in the numbers as well. Synthesia processed 12,450 user appeals in 2025. Around 31% were approved on review and the original decision was reversed, while around 69% were upheld. This is the distribution expected from a system that catches genuine violations consistently while still leaving room for legitimate use cases to be reinstated when a classifier gets a call wrong.
Proof under pressure
Leadership when it matters most
The stakes were illustrated when Belgian public broadcaster RTBF aired its investigation IA, la fabrique à arnaques (AI, the scam factory). As part of the broadcast, RTBF tested AI video tools to see which would generate the content fuelling scams and other harms. Synthesia blocked the creation of non-consensual deepfakes, political propaganda, racist and xenophobic content, and crypto investment scams. Several other AI video platforms generated all of them.
That contrast is the reason Synthesia has built the company the way it has, and the reason it is now partnering with Cinder. Enterprise customers depend on the platform because they know it will refuse to be a weapon — and the wider public depends on it too, because the standards an industry sets in its first years are the ones it lives with for a long time afterwards.
What comes next
A safer Synthesia is good for more than Synthesia.
AI video is one of the fastest-growing surfaces on the internet, and the choices the industry makes now about where moderation sits, who is accountable, and what gets reported transparently to the public will shape the category for years.
With Cinder as a partner, the Synthesia trust and safety team can build faster against new harm types, validate decisions against richer signals, and stay ahead of regulatory regimes as they evolve.
Watch
See Synthesia's integrity in action
Build on the safest foundation
Talk with the team helping Synthesia keep enterprise AI video safe — moderation infrastructure that moves at the speed of your roadmap, not behind it.