OpenAI and TikTok sign new pledge to watermark AI content

Generative AI moves at breakneck speed, and formal regulations are still struggling to catch up. In the meantime, big tech is trying out the honor system.

A coalition of ten companies—including OpenAI, TikTok, Adobe, the BBC, and dating app Bumble—just signed onto a voluntary framework to build, create, and share AI-generated content responsibly. The guidelines were drafted by the Partnership on AI (PAI), an AI research nonprofit that consulted with over 50 organizations.

The initial signers also include Canadian broadcaster CBC/Radio-Canada, human rights nonprofit Witness, and synthetic media startups Synthesia, D-ID, and Respeecher.

Tagging the synthetic wave

The core of the agreement boils down to transparency.

The framework calls on both AI developers like OpenAI and distribution platforms like TikTok and the BBC to be upfront about what their tools can and cannot do. Crucially, they agreed to tell users whenever they are interacting with synthetic media.

In practice, that means testing and deploying watermarks, visual disclaimers, and traceable metadata inside model training data.

Claire Leibowicz, head of AI and media integrity at PAI, noted that the goal is to keep synthetic media from being used to harm or disenfranchise people, and instead steer it toward creativity and knowledge sharing. Henry Ajder, a generative AI expert who helped draft the guidelines, described the current landscape as a "Wild West" and hopes the pledge gives businesses a baseline roadmap.

The problem with the honor system

Not everyone is convinced that a voluntary pledge will make a real dent.

Hany Farid, a UC Berkeley professor studying deepfakes, pointed out that voluntary industry guidelines rarely work in practice.

The main issue is simple: even if OpenAI puts guardrails on ChatGPT and DALL-E, companies outside the pact are under no obligation to comply. Open-source models like Stability AI’s Stable Diffusion, for instance, can easily be run without restrictions, letting anyone generate inappropriate images or nonconsensual deepfakes.

Farid argues that real protection requires hard mandates. He wants app stores and cloud providers run by Amazon, Microsoft, Google, and Apple to actively ban tools used to create nonconsensual sexual imagery, while making watermarking mandatory across the board.

Intel senior research scientist Ilke Demir and Access Now analyst Daniel Leufer also noted major omissions inside the guidelines. While the text targets fraud, harassment, and disinformation, it skips model-level issues like toxic training data and demographic bias—such as image models that default exclusively to white faces. Farid posed an even more fundamental question: if tech companies know the harms are this severe, why aren't they asking whether they should build this tech in the first place?

Why it matters

For everyday internet users, this pledge might soon mean spotting clearer "AI-generated" tags and watermarks across TikTok clips, BBC coverage, or dating profiles on Bumble. That offers a helpful layer of context, but given how easy it is to spin up unconstrained open-source models, users still can't let their guard down.

For the tech industry, this is classic pre-emptive self-regulation. With the European Union pushing to fold strict synthetic media disclosures into its upcoming AI Act, signing a voluntary pledge lets major players show regulators they can police themselves before heavy fines enter the picture.

We think getting model builders, creative toolmakers, and social platforms aligned on watermarking is a smart and necessary baseline.

However, the lack of any enforcement mechanism is an obvious weak spot. When there are zero penalties for skipping disclaimers, a voluntary pact signed by ten well-behaved companies won't stop bad actors from flooding the web with synthetic chaos.