The Internet’s New Authenticity Crisis

A New Authenticity Crisis Is Spreading Online

The internet’s long-running problem with fakery is entering a new phase: the tools for creating convincing synthetic media are becoming cheaper, easier and more deeply embedded in the largest platforms, even as those same companies race to contain the fallout.

In recent days, Google expanded consumer video-generation features that let people create lifelike avatars of themselves, while LinkedIn said it was stepping up efforts to limit what it calls inauthentic activity, including automated comments and suspiciously amplified posts. At the same time, new reporting and research have underscored how quickly low-cost synthetic content is proliferating across social feeds — from monetized hate propaganda on Facebook to AI-generated faces that ordinary users struggle to identify any better than chance.

Taken together, the developments point to a widening crisis of authenticity online. The basic question facing users — is this real? — is becoming harder to answer, just as platforms make synthetic content easier to produce at scale.

Google Pushes AI Video Further Into the Mainstream

At its I/O conference this week, Google introduced a broader set of AI video tools across Gemini, Flow and YouTube Shorts. Among the most striking additions was a feature allowing users to create a custom AI avatar that resembles and sounds like them, effectively lowering the barrier to making a persuasive selfie-style video without filming one in the traditional sense.

Google has presented these tools as creative utilities, part of a larger push to make AI-generated video a mainstream consumer format. YouTube says Shorts made with its own AI tools are automatically labeled, and the company has also built a likeness-detection system intended to help identify possible deepfakes of creators and public figures.

But the easier it becomes to generate a realistic version of oneself, the more difficult it may be for viewers to know when they are watching a genuine recording, a stylized recreation or something more deceptive. What once required technical expertise can now be done inside products used by hundreds of millions of people.

Platforms Try to Clean Up What They Helped Create

That tension is becoming increasingly visible across the industry. On one side, technology companies are promoting generative AI as a new layer for everyday communication. On the other, they are trying to reassure users that their platforms are not drowning in machine-made junk.

LinkedIn is the latest example. The professional networking site said it would curb the reach of inauthentic activity, including spamlike automation and engagement manipulation, while broadening profile and page verification. The company has said early systems to identify generic AI content have shown high accuracy rates, and it has leaned heavily on verification as a trust signal; more than 100 million members, LinkedIn said earlier this year, have added at least one verification.

The move amounts to a tacit acknowledgment of a problem many users have already noticed: feeds increasingly crowded with polished but empty posts, formulaic comments and synthetic engagement. The irony is that LinkedIn’s parent company, Microsoft, has been one of the loudest corporate champions of AI-powered productivity and content tools.

The contradiction is not unique to LinkedIn. Across the industry, platforms are encouraging users to create with AI while also trying to distinguish acceptable assistance from deceptive or low-value spam.

The Darker Side of “AI Slop”

The consequences go beyond annoying posts.

An investigation by The Bureau of Investigative Journalism and The Guardian traced networks of Facebook pages pushing hateful, AI-generated content about Britain to operators in South Asia, including Sri Lanka and Pakistan. The material reportedly mixed anti-immigrant and anti-Muslim themes with manufactured nostalgia and inflammatory memes, and was designed not only to provoke but also to generate revenue.

Meta removed some pages after being contacted, according to the reporting, but researchers found that replacement networks could quickly reappear. That pattern reflects a familiar challenge in platform moderation: when the cost of producing content collapses, enforcement becomes less like removing a few bad actors and more like trying to contain a constantly regenerating supply chain.

The significance of such content is not merely that it is fake. It is that synthetic media can now be used to industrialize persuasion — and resentment. A post no longer needs to be handcrafted to be politically effective; it only needs to be good enough to trigger clicks, shares and outrage.

People Are Worse at Spotting Fakes Than They Think

If platforms hope users will act as a last line of defense, new research suggests that confidence may be misplaced.

Researchers at the University of New South Wales recently found that people perform only around chance when asked to distinguish AI-generated faces from real ones, despite often expressing confidence in their judgments. That finding reinforces what many internet users have already sensed: visual cues that once signaled fakery — odd eyes, warped backgrounds, unnatural skin — are becoming less reliable.

As image and video models improve, the burden of detection falls on audiences who are poorly equipped to carry it. The result is not just that more people may be fooled. It is that everyone may become less certain about everything they see.

That erosion of confidence carries broader costs. Authentic creators can be doubted. Real evidence can be dismissed as fabricated. Fraudsters and propagandists gain room to operate in the fog.

A Search for Technical Proof

In response, AI companies are increasingly embracing provenance tools — digital signals meant to show how a piece of media was created and whether it has been altered.

OpenAI said this week that it was expanding support for content provenance by combining C2PA Content Credentials, Google’s SynthID watermarking and a public verification tool for images made with its systems. The effort reflects a growing industry view that origin signals need to become basic infrastructure for online media, not an optional extra.

Yet even the companies promoting these systems acknowledge their limits. Provenance markers can be stripped away through screenshots, cropping, reposting or platform handling. Detection systems are imperfect. Labels can help, but only if they remain attached to the content as it spreads.

That leaves the internet in an awkward transitional period: AI-generated media is becoming frictionless to make, while the mechanisms for proving origin remain fragile and unevenly adopted.

Why This Matters Now

For years, concern about deepfakes centered on spectacular threats — a fake politician’s speech, a fabricated wartime video, a celebrity impersonation. Those dangers remain. But the more immediate problem may be more mundane: a steady flooding of feeds with synthetic faces, synthetic voices, synthetic clips and synthetic sentiment.

When realistic self-avatars, generated videos, fake profile pictures and mass-produced propaganda all become abundant, trust does not break in one dramatic moment. It degrades by increments. Users hesitate a little more. Platforms rely more heavily on verification and ranking systems. Bad actors gain more tools to impersonate, mislead and manipulate.

The fight over AI slop, in other words, is not really about aesthetics. It is about whether the social web can continue to function when the cost of manufacturing plausibility approaches zero.

Sources

Further reading and reporting used to add context:

AI News