This is a cautionary story. It's fresh, it's unresolved, and if you're working with AI audio or dubbing workflows, it's directly relevant to you.

Our CTO's YouTube account was permanently banned while running internal tests on AI speech-to-speech dubbing. Legitimate work, supporting a real client brief around multilingual dubbing. Just a STS engine (Gemini by the way) doing what it's designed to do.

The formal appeal was submitted and ban decision subsequently upheld by YouTube. We're continuing to seek resolution through other channels, but the account remains suspended. It's a frustrating outcome, and one that raises questions we think are worth putting to the wider industry.

How It Happened

A client came to us wanting to explore speech-to-speech dubbing: translating spoken content into another language while keeping the character and feel of the original voice. It's a capability that's genuinely in demand right now, as more organisations look to reach multilingual audiences at scale.

To see how well our tooling performed, our CTO ran an internal test using a recording featuring a celebrity figure. The engine? Gemini, one of the AI engines available through our platform, Line 21. The goal was purely linguistic. Translate the speaker into another language. That's it.

YouTube classified it as a synthetic content violation and issued a permanent ban.

The Policy Collision Nobody Is Talking About

Here's where things get complicated, and more than a little ironic.

YouTube has its own automatic dubbing feature. It supports expressive speech-to-speech translation across dozens of language pairs: English to Spanish, French, German, Hindi, and many more. The technology YouTube is penalising others for testing is, in a meaningful sense, the same category of technology YouTube itself is actively building and rolling out.

YouTube's GenAI disclosure policy does require creators to flag content that "makes a real person appear to say or do something they didn't do." That's a reasonable principle. But automated enforcement appears broad enough to catch legitimate translation workflows, particularly when a recognisable voice is involved.

YouTube's privacy guidelines go further, allowing removal requests for AI-generated synthetic content that simulates a real person's likeness, with extra scrutiny on content featuring public figures.

What this creates, in practice, is a policy environment where automated systems struggle to tell the difference between language translation and identity manipulation. That's a significant gap.

Why This Matters for AV Teams and Broadcast Professionals

If you work in live captioning, translation, or AI-assisted dubbing, this isn't just a platform policy curiosity. It has real operational implications for how you test, demo, and deliver work.

Consider what's now potentially at risk:

• Internal QA testing of speech-to-speech engines using real-world audio samples
• Client demos featuring dubbed content with well-known speakers
• Broadcast translation workflows where the original speaker's voice is carried into another language

Any of these could, under current enforcement logic, trigger a synthetic content flag. Not because the intent is deceptive, but because the detection systems aren't yet calibrated to distinguish dubbing from deepfaking.

For teams running regular event services where uptime and uninterrupted delivery are non-negotiable, a sudden platform ban mid-workflow isn't just disruptive. It can be catastrophic.

What You Should Be Doing Now

This isn't a reason to stop developing or testing AI dubbing capabilities. The technology is moving fast and client demand is real. But there are some practical steps worth building into your process:

1. Review your disclosure practices. If you're uploading AI-dubbed or translated content to YouTube, even for internal testing, check whether it meets YouTube's disclosure requirements. Label it clearly as AI-generated or synthetically altered.

2. Avoid using identifiable public figures in test content. This is the most direct way to reduce exposure to automated flagging, at least until platform policy catches up with legitimate use cases.

3. Brief your clients. If you're offering speech-to-speech dubbing and clients plan to publish that content on YouTube, they need to understand the current enforcement landscape before delivery, not after.

4. Document your workflows. If a ban or flag does occur, clear records of intent, methodology, and tools used will support any appeal. In our case, the appeal was upheld and the ban remains in place, though we are still seeking resolution through other channels.

5. Watch for policy updates. YouTube's synthetic content policies are evolving. What's ambiguous today may be clarified, or tightened, within months.

A Question for the Community

We're sharing this because we doubt we're the first.

Has your team run into incorrect flagging or platform action while working with AI dubbing or speech-to-speech tools? Have you had to adjust your internal testing practices, or brief clients on this risk?

We'd genuinely like to hear from AV professionals, broadcast engineers, and media technology teams working in this space. The more visibility we have as an industry, the better placed we are to respond, whether through collective feedback to platforms, updated working practices, or simply knowing what to look out for.

Platform policies around synthetic content are tightening, and for understandable reasons. But if enforcement systems can't tell a dubbed translation from a deepfake, that's a problem the industry needs to surface clearly and soon.