We are 'dangerously close' to a dead internet, says one AI-detection startup

Key points

• AI-generated text and images now appear regularly in job applications, product reviews, and insurance claims, according to reporting by TechCrunch AI.
• The CEO of Pangram, an AI-detection company, says the internet is "dangerously close" to what theorists call the "dead internet": a web where most content is machine-made, not human.
• A small number of startups have launched AI-detection tools in the past two years, but none has published peer-reviewed evidence of accuracy at scale.
• Ordinary people encounter this problem every day, whether they are a hiring manager reading a cover letter or a shopper weighing up reviews.

What is the "dead internet" and why does it matter now?

The "dead internet" theory holds that most online content is already created by bots, not people. It started as a fringe idea; it is starting to look like a reasonable concern.

For years the theory lived in the corners of online forums. The argument was simple: if automated accounts could generate content cheaply and at scale, why would anyone pay humans to do it? Now that large language models, the technology behind chatbots like ChatGPT, can write a convincing cover letter in seconds, the economics have shifted sharply.

Pangram's chief executive put it bluntly: the internet is "dangerously close" to that tipping point.

Where is fake AI content actually showing up?

Everywhere incentives exist to cut corners. Job seekers use AI to bulk-apply for roles, producing polished letters that tell a hiring manager nothing true about the applicant. Sellers flood e-commerce sites with AI-written five-star reviews. And, more troublingly, people submit AI-generated supporting documents in insurance claims, which is fraud.

Each of these cases harms a real person at the other end: the honest applicant passed over, the shopper who buys a dud product, the insurer, or, ultimately, every policyholder who pays higher premiums.

Can detection tools actually fix this?

Not reliably, not yet. A handful of startups have built tools that claim to flag AI-generated text or images. But none, as of publication, has shared peer-reviewed results showing their accuracy holds up across different types of content and different AI models. Company claims and independently verified results are different things, and right now we only have the former.

Detection is also a moving target. The AI models that generate content improve constantly. A detector trained on today's output may miss tomorrow's.

Problem area	Who it harms	Current fix
AI job applications	Honest candidates, employers	Verbal interviews, skills tests
Fake product reviews	Shoppers	Platform moderation (inconsistent)
Fraudulent insurance claims	Insurers, policyholders	Forensic document checks
General web content	Everyone	No reliable fix yet

What should ordinary people do?

Treat surprisingly polished, frictionless text with mild scepticism. If a review says exactly what you want to hear with no rough edges, that is worth noticing. Employers may want to move more weight onto structured interviews and practical tasks rather than written applications alone.

No single tool will solve this. The responsibility sits partly with platforms to invest in moderation, partly with regulators, and partly with the companies building the generative AI that makes cheap faking so easy.

Common questions

How accurate are AI-detection tools right now?

No AI-detection company has yet published independent, peer-reviewed accuracy figures that hold up across different content types. Treat any detection score as a useful signal, not a verdict.

Is using AI to write a job application illegal?

Generally no, but using AI to fabricate supporting documents in an insurance claim is fraud. The legal and ethical lines vary by context.

Will this get worse before it gets better?

Detection research is running behind generation research, so in the short term, yes, spotting AI-made content is likely to get harder, not easier.