Could AI kill the internet? Synthetic content and the credibility crisis
A short, sharp Reddit post titled “AI could kill the internet” has struck a nerve. The claim is simple: if anyone can generate convincing text, images, audio, or video at scale, the web becomes untrustworthy by default.
It will soon get to the point where everything on the internet can’t be trusted to be real.
You can read the original discussion here: AI could kill the internet.
Is this hyperbole, or a fair warning? There’s truth in both directions. Synthetic content is cheap and fast. Trolls and scammers will use it. But there is also a realistic path to restoring trust: provenance, transparency, and platform-level changes that reward verified sources.
What the Reddit post gets right: incentives and scale
Generative models make it trivial to produce persuasive content. That shifts the cost curve in favour of misinformation, spam, and harassment. It also amplifies the “liar’s dividend” – when convincing fakes exist, bad actors can dismiss inconvenient truths as fake, too.
For UK readers, the practical risks are clear:
- Fraud and scams, especially deepfake voice and WhatsApp-style impersonation.
- Information disorders around public events, elections, or breaking news.
- Corporate risks: reputational damage, phishing, market manipulation, and internal data leakage.
Can detectors and watermarks save us?
Detection is part of the answer but not the whole solution. AI “detectors” try to spot generated content; watermarking tries to embed hidden signals in outputs. Both approaches are helpful, but neither is bulletproof. Models evolve, content is edited, and signals can degrade or be stripped.
If you want to see how watermarking works in principle, Google’s SynthID is a representative effort: it embeds signals into media that aim to persist through common edits. See the overview on Google DeepMind’s SynthID page. Results and robustness vary by modality and use case.
The path to provenance: cryptographic signatures and content credentials
What looks more durable is provenance – recording where content came from and how it changed. The leading cross-industry standard here is C2PA (Coalition for Content Provenance and Authenticity). It attaches cryptographically verifiable “Content Credentials” to media, showing capture device, edits, tools used, and authorship when disclosed.
- C2PA – an open standard for signing media and edit histories.
- Content Credentials – user-facing way to view that provenance.
Provenance won’t stop bad actors creating fakes, but it helps good actors prove their work is authentic. Over time, platforms and search engines can prioritise signed, provenance-rich content, much as browsers prioritise HTTPS over plain HTTP.
What UK organisations and teams can do now
For individuals
- Check the source. Prefer original reporting and official accounts over screenshots and reposts.
- Use basic forensics: reverse image search, time-stamped archives, cross-reference quotes.
- Look for Content Credentials when available, and treat unlabelled viral media with caution.
- Be sceptical of urgent money or data requests, especially via voice notes or calls.
For product teams and comms
- Adopt C2PA signing for images, audio, and video you publish. Document your edit pipeline.
- Watermark AI outputs where possible and disclose when content is synthetic or edited.
- Build citation into generative features. Retrieval-augmented generation (RAG) – where a model quotes and links to a trusted knowledge base – improves transparency.
- Maintain an audit trail of prompts, models, and outputs for compliance and incident response. Even a simple logged workflow can help; for example, structuring outputs into Sheets or a database so you can trace decisions later. I’ve shown one approach here: Connect ChatGPT and Google Sheets.
- Rate-limit and review user-generated content. Expect a flood of synthetic submissions.
For UK compliance and governance
- Align with UK data protection guidance when using generative AI, especially for personal data.
- Prepare for platform and regulatory shifts under the Online Safety regime and related codes. Expect duties around harmful content and fraud mitigation.
- Publish a clear AI use policy: where you use AI, what models, how you mitigate bias and errors, and how users can report issues.
What to watch next: platforms, policies, and norms
Trust won’t be rebuilt by individuals alone. Expect changes in:
- Search and social ranking – giving preference to verified sources and provenance-rich media.
- Creator tools – cameras, phones, and editing apps that sign content by default via C2PA.
- Model vendors – expanding watermarking and disclosure tools across text, image, and audio.
- Newsrooms and public sector – adopting provenance to protect public information and counter hoaxes.
Bottom line: AI won’t “kill” the internet, but it will force a trust reset
The Reddit post captures a real risk: synthetic content at scale can collapse credibility. But the response is already forming. Provenance standards, platform incentives, and better user habits can make authentic content easier to verify than fakes are to deny.
The near future of the web is a layered one: more AI, more synthetic media, and a parallel system that shows what’s real, how it was made, and who stands behind it. That’s not the end of the internet – it’s an overdue upgrade.
