Whether it is poor quality content or dealing with faked videos of both current and historical figures, broadcasting is not alone in facing challenges from AI generated content. Such content can cause very real reputational damage, which is why broadcasters are investing time and effort in meeting these challenges head on – especially when it comes to collaborations with third-party platforms and content distributors.

Increasingly beholden to the major social platforms for enforcing rules over faked images and AI-generated content, these challenges further highlight broadcaster’s loss of control over how content reaches their audience. This is particularly the case with YouTube, with whom a growing number of broadcasters are collaborating. The BBC announced a strategic partnership with YouTube in January 2026 aiming to complement its iPlayer and BBC Sounds services to increase its audience reach, but also to cut through with high quality authenticated news.

Just three months later, in April 2026, France Televisions unveiled a similar collaboration with YouTube as part of its "streaming first" strategy, making thousands of hours of programs accessible each year. Again, its primary objective is to increase the reach and reputation of its news content.

This plays directly into the issue of AI slop.

AI Slop

The Cambridge Dictionary definition of “slop” was amended in June 2025 to account for the use of generative AI and is now also defined as “content on the internet that is of very low quality, especially when it is created by artificial intelligence.”

It is a timely addition, because broadcasters and a growing number of other content providers now rely for a significant proportion of their distribution on platforms heavily polluted with so-called AI slop. According to a report in the Guardian newspaper in December 2025, more than 20% of content shown to freshly opened YouTube accounts is now "low-quality AI-generated content designed to farm views".

Broadcasters might calculate this gives them an opportunity to differentiate themselves from all the dross around them; there is already evidence of a backlash against low quality AI generated content, including fakes or simulations of past experts or celebrities pontificating on any plausible subject. Such material is often given away not so much by the video, or even the nature of the content, but by its poor-quality audio, which exposes a lack of sound expertise among creators. This also presents an opportunity for broadcasters to generate higher quality AI content themselves, which some seem to be doing.

A Broadcaster Backlash

But even higher-quality AI content has risks, as the BBC discovered in May 2026 when it featured an AI-generated panel comprising historical figures for its topical show Question Time, which allows members of the public to interrogate leading politicians and media commentators. This panel included UK second world war leader Winston Churchill, pioneering artist Frida Kahlo, as well as suffragette Emmeline Pankhurst and leader of India’s independence movement Mahatma Gandhi.

The program drew flak even before it went on air after a clip was posted on the show’s social media site, with criticisms over both ethics and the program’s environmental impact. But the bigger issue was the BBC’s apparent endorsement of faked AI images after it had heavily criticized social media firms over their pivot towards such practices in one of its documentaries. The documentary accused them of peddling AI slop, with particular ire for Meta’s Mark Zuckerberg after he had declared AI generation as the third wave of social media in October 2025, with friends and family being first and creator content second.

AI Watermarking

Lingering behind such rhetoric is a hope on the part of broadcasters that the platforms themselves will be provoked by the criticisms into bearing down harder on AI slop.

There is some sign of that happening, with Google first to take a significant step with the launch of AI based watermarking early in 2024 with its SynthID tool. This is similar in concept to forensic video watermarking used to identify video streams and film content, mandated by MovieLabs in 2014. Google says Synth ID has already been used to watermark over 100 billion images and videos and been adopted by other leading tech firms, such as Nvidia, Kakao and ElevenLabs.

AI watermarking is designed to embed marks imperceptibly so that they are invisible to the user and yet be readily detectable and resistant to tampering or efforts to remove them. Such AI watermarking is also being combined with blockchain technology to help ensure tamper-proof verification. Blockchain enables the creation of immutable ledger records that allow watermark data to be distributed across multiple decentralized networks, which makes it much harder to alter or forge content without detection.

Indeed, a key objective of many contemporary security systems is to accept that violations occur, but ensure that when they do they are detected, thereby minimizing or eliminating the threat. That is the idea with quantum cryptography, for example.

In May 2026 OpenAI announced that in future all images generated by ChatGPT, Codex, and the OpenAI API would carry a Synth ID watermark. They were already carrying the C2PA manifest from the Coalition for Content Provenance and Authenticity, which gives more detail on the origins and history of the content. There is a synergy between the two and Google also supports C2PA, and we will go into detail about how they work in part 2 of this miniseries.

Another aspect of AI watermarking which is of interest to platforms and distributors of third-party content its potential to reduce the cost of content moderation, which is a fast-moving target. AI watermarking can automate detection of deepfakes, while focusing resources on high-risk content more likely to require moderation. This is also relevant for regulators, broadcasters, content creators and any party threatened by misinformation and faked material.

It is going to be an evolving landscape involving an arms race with attackers. The experience of forensic watermarking has shown that the more effective a tool is, the greater attention it attracts from hackers, cyber criminals and other bad actors.

But there is confidence that AI watermarking represents a promising avenue for combating AI fakery and fraud, with scope for further innovations such as resistance against quantum computer attacks on conventional cryptographic systems, which again we discuss in part 2.

As a result of this potential, a growing number of regulators and governments are making it mandatory to embed standardized watermarks in images, video and text generated by AI algorithms. Both the EU and US have introduced regulations to enforce such compliance, initially for applications deemed high risk, such as political elections and healthcare. It is still early days as standards are only emerging, as in that coalescence around Google’s SynthID.

Legislation

Legislators are still struggling over how to regulate AI content generation, in part because it is such a fast-evolving field that the models continue to advance rapidly in capability. To some extent they are making it up as they go along, as is the EU with its AI Act, displaying the same ambition to establish a global precedent as it did back in 2016 when it enacted the rather contentious GDPR privacy framework, which took effect in May 2018.

So far, the EU AI Act has elicited more confusion than compliance, but that is no worse than other jurisdictions. It runs in parallel with the EU Digital Services Act (DSA), which requires the biggest online platforms to proactively mark deepfakes distributed on their platforms. The AI Act similarly obliges all deployers to disclose that image, video, or audio deepfakes are AI-generated. In order to determine when the Act applies, the EU defines a deepfake as an AI-generated or manipulated image, audio or video content that resembles real people or objects and gives a false appearance of authenticity.

The EU has gone some way to balance creative freedom with protection of image rights and privacy. It now allows deepfakes that may be deceptive but are not illegal under a given member state’s national law to be merely labelled, not deleted. Furthermore, the extent of the labelling can be proportional to the content type, so as not to be too intrusive to the viewer or listener. The AI Act allows artistic, creative, satirical, fictional or analogous content to be labelled in “an appropriate manner” that “does not hamper the display or enjoyment of the work.”

Establishing compliance is technically complex and beyond the capabilities of many smaller content creators. For this reason, the EU is leaning on the big platforms, which are all capable of meeting the requirements. The DSA imposes a tiered system of obligations, with the most stringent requirements only applying to platforms with over 45 million monthly EU users. For these so-called VLOPs (Very Large Online Platforms), annual assessments are required to identify significant systemic risks stemming from their services. These include dissemination of illegal content, negative effects on fundamental rights such as human dignity and privacy, and adverse impacts on civic discourse.

Sports

There are also issues arising from other spheres, especially sports, where image rights can be highly valuable and at risk of being infringed by AI generation. Among numerous cases involving sporting superstars, especially in football, was an AI generated image of David Beckham dancing while wearing the England three lions’ kit. Few people were fooled given that the image appeared more than 15 years after he last played for England, but such cases have led to growing pressure on the platforms to take stronger action to take down such images.

Some platforms have set up quasi-independent bodies to arbitrate such cases, notably Meta, which in 2018 established its Oversight board, which now appoints its own members to maintain independence. It has banned various AI generated ads that have infringed image rights, including one imitating the voice of Brazil striker Ronaldo. This slipped through Meta’s automatic detection filters, but led to stronger checks over simulated audio, so was an example of a major platform raising the protective bar further in response to an infringement and associated pressure.

Many broadcasters will be familiar with such pressure from their experiences distributing live premium sport where they may be reliant on techniques such as forensic watermarking to protect against piracy, especially illicit stream redistribution. They are therefore watching the unfolding situation over AI generation, where again they may find themselves in the crossfire between players, leagues, advertisers, platforms and users.

There is also a new category of company offering to look after the rights of both clubs or teams, and players. Such firms in turn use AI to identify when a company or individual’s image rights or intellectual property has been infringed, and then issue takedown notices, without the parties affected necessarily being directly involved at all.

This again has echoes in traditional forensic video watermarking, and we will cover more technical detail on that in part 2.