Joined February 2019
1,917 Photos and videos
Pinned Tweet
EU AI Act Article 50 enforcement begins in August. Content creators and platforms in the EU will be legally required to label AI-generated output. The question is not whether to comply. It is what compliant labeling actually looks like under audit. A label in a caption is a claim. A C2PA manifest signed at origin is a record. One can be deleted or disputed. The other cannot. Numbers Protocol implements C2PA at the infrastructure layer. Every asset registered through the Capture SDK carries a signed manifest and an on-chain record on Numbers Mainnet. The record exists independent of any platform that displays the content. Enterprise newsrooms are already running this infrastructure in production. August is eight weeks away. The infrastructure is already running. docs.captureapp.xyz/capture-…
2
6
29
688
most compliance solutions are labels applied after the fact. provenance is different. it is attached at the moment of creation. you cannot fake a timestamp you did not control. that is the entire product.
China Times photographers register images at the moment of capture, before any editorial process begins. Shoot, upload through the Capture SDK, receive a NID. The C2PA manifest is embedded in the asset. The record lands on Numbers Mainnet. At any point after publication, the provenance chain is independently verifiable. No reliance on the publication platform. No reliance on internal records that can be edited. The record was made at origin. That is what makes it auditable. EU AI Act enforcement begins in August. Newsrooms have eight weeks. This is what a compliant pipeline looks like for a newsroom that built the receipt into the workflow, not bolted it on after. x402-chinatimes.numbersproto… Receipts before deadline.
2
3
12
404
A photo can be generated, edited, and posted before anyone thinks to ask where it came from. The asking always happens too late. So the answer has to exist before the question. Recorded at the moment of creation, not reconstructed after the fact. That is the part a label cannot do.
the EU spent 7 months and hundreds of experts writing the rulebook for AI content labels. their own conclusion: no single watermarking technique is reliable enough. you need layers. final code published this week. Article 50 applies 2 August, 52 days to go. a label slapped on at publish can be stripped, cropped, or screenshot away. a record made at capture survives all three. ProofSnap does the capture part. the part the watermark can't.
14
481
the EU decided AI content needs receipts. Article 50 brings machine-readable marking, detection, and disclosure duties for AI-generated and manipulated content. most teams are still treating this like a label problem. the ones who will be ready are the ones who build the record before the mandate. provenance attached at capture is not the same as a label applied at publish.
the EU spent 7 months and hundreds of experts writing the rulebook for AI content labels. their own conclusion: no single watermarking technique is reliable enough. you need layers. final code published this week. Article 50 applies 2 August, 52 days to go. a label slapped on at publish can be stripped, cropped, or screenshot away. a record made at capture survives all three. ProofSnap does the capture part. the part the watermark can't.
3
3
21
525
Everyone is still arguing about whether AI content should be labeled. The law already decided that. It starts in August. The open question is quieter. Who can actually prove what their model made, after the fact, without taking their word for it. That part only gets solved by a record, not a label.
the EU spent 7 months and hundreds of experts writing the rulebook for AI content labels. their own conclusion: no single watermarking technique is reliable enough. you need layers. final code published this week. Article 50 applies 2 August, 52 days to go. a label slapped on at publish can be stripped, cropped, or screenshot away. a record made at capture survives all three. ProofSnap does the capture part. the part the watermark can't.
5
5
23
1,087
Hi Numbers Community! A caption is just a claim. However, a C2PA record on Numbers Mainnet is the proof. EU final Code of Practice on AI content transparency was published this week. Article 50 enforcement will start on August 2, 52 days remaining for platforms to comply. Read the full letter: docs.numbersprotocol.io/intr…
2
4
17
396
The audit trail does not start at publish. It starts when the image, article, or agent output is created. If the record begins only after the dispute, it is just a story with better formatting. Human Truth. Machine Proof.
China Times photographers do not upload first and register later. They register at origin. Before any image enters the editorial pipeline, it carries a NID, a C2PA manifest, and an immutable record on Numbers Mainnet. This is done for the sake of professional standard not as compliance checkbox. The EU AI Act changes what "professional standard" means for AI-generated content. Enforcement begins in August. Traceability is no longer optional and will be a liability question. Most AI content pipelines produce no chain of custody at all. That gap does not appear on a balance sheet. It appears in a compliance audit or a legal proceeding where there is nothing to present. The Capture SDK closes that gap behind one API call. The record exists before the content leaves the source. Build it in now, or retrofit it later under pressure.
2
4
15
620
the EU spent 7 months and hundreds of experts writing the rulebook for AI content labels. their own conclusion: no single watermarking technique is reliable enough. you need layers. final code published this week. Article 50 applies 2 August, 52 days to go. a label slapped on at publish can be stripped, cropped, or screenshot away. a record made at capture survives all three. ProofSnap does the capture part. the part the watermark can't.
2
3
21
2,617
China Times photographers register images at the moment of capture, before any editorial process begins. Shoot, upload through the Capture SDK, receive a NID. The C2PA manifest is embedded in the asset. The record lands on Numbers Mainnet. At any point after publication, the provenance chain is independently verifiable. No reliance on the publication platform. No reliance on internal records that can be edited. The record was made at origin. That is what makes it auditable. EU AI Act enforcement begins in August. Newsrooms have eight weeks. This is what a compliant pipeline looks like for a newsroom that built the receipt into the workflow, not bolted it on after. x402-chinatimes.numbersproto… Receipts before deadline.
2
3
16
852
Digital media has two jobs now. Stay private when it is shared. Stay verifiable when it is trusted. Numbers Protocol is partnering with @limewire to cover both. Private files. Verifiable provenance. That is what secure media should look like.
3
11
39
3,234
EU AI Act Article 50 requires a machine-readable disclosure for AI-generated content. C2PA provides it. A C2PA manifest is a JSON-LD assertion embedded in the file at origin, cryptographically signed by the creating tool or service. It contains: who created it, what tool was used, when it was created, and whether it has been modified. The challenge is provenance at scale. A manifest embedded in a JPEG can be stripped. The on-chain record on Numbers Mainnet cannot. The Capture SDK creates both in one call. The C2PA manifest goes into the asset. The NID and timestamp go on-chain. Both are independently verifiable without access to the original file. PyroImage uses this stack for photojournalism provenance across publication pipelines. This is what Article 50 compliance looks like as infrastructure rather than a checkbox. x402-pyro.numbersprotocol.io
2
4
19
473
you can add an "AI-generated" label to a real photo. you can remove it from a fake one. the label everyone is about to mandate can be faked in both directions.
5
4
17
532
a provenance record is different. who made it. when. which tool. whether it changed since, attached at creation. that is the part we built.
1
8
210
the law says content must be labeled. it does not say the label has to be true. that gap closes eventually. august is closer than it looks.
1
1
9
195
The POAP Campaign with @acurast is now open. Last week we spent today's X Space talking about what provenance infrastructure means in practice. this is the practice part. Acurast runs decentralized compute. every job it runs can now have an on-chain proof of what happened. Claim your POAP. free. agents and humans both eligible. Five wallets win 8,000 NUM each. ama.numbersprotocol.io/ama/2…
5
4
19
543
The Olympex AMA POAP campaign closes today. Thank you to everyone who participated: humans and agents. Five winners take home 8,000 NUM each. Results below. A new POAP campaign coming up in 7 hours
2
4
20
532
We're having an X Spaces with @exolix_com on Jun 9 at 14:00 UTC. $NUM is listed on Exolix. This session covers what provenance infrastructure means beyond a token listing, how Numbers Mainnet records every content registration on-chain, and what comes next for agent accountability. Set your reminder here: x.com/i/spaces/1aJbddyNvjXKX
🎙️ Join us for an AMA with @numbersprotocol on Jun 9 at 14:00 UTC. We'll discuss content provenance in the AI era, the $NUM ecosystem, and the future of verifiable digital content. 🔔 Set a reminder: x.com/i/spaces/1aJbddyNvjXKX
1
17
654
Numbers Protocol retweeted
🎙️ Join us for an AMA with @numbersprotocol on Jun 9 at 14:00 UTC. We'll discuss content provenance in the AI era, the $NUM ecosystem, and the future of verifiable digital content. 🔔 Set a reminder: x.com/i/spaces/1aJbddyNvjXKX
8
8
22
1,697