AI-generated CSAM in H1 2026 topped all of 2025
The safety data and the identity-verification push are answering the same question from opposite ends: how do you tell real people from generated ones? Three stories worth your midday.
AI-generated child sexual abuse material is compounding faster than any detection number can normalize it. The Internet Watch Foundation assessed 6,310 AI-generated images that meet the legal definition of CSAM in the first six months of 2026 — about 40% more than the 4,512 it assessed in all of 2025, and that full-year figure was itself a record. The composition is what should worry policymakers more than the headline count: 98% of images with a recorded gender depicted girls, 79% showed children aged 7 to 13 (up from 70% last year), and 88% fell into the most severe legal category, Category C, versus 62% in 2025. The IWF is careful to say these are still images only — video is counted separately — and it is explicitly arming its argument with the numbers: its policy team is calling for binding AI legislation, and its case now rests on a doubling curve that better moderation cannot outpace. Generation cost is collapsing while the harmful end of the market scales like a franchise; detection reports are the lagging indicator, and they are already behind.
World is bringing proof-of-human verification to Zoom — because a coin flip now decides whether you trust your caller. World's Zoom integration matches three things cryptographically: the signed image from a person's original Orb verification, a fresh selfie, and the live video frame, then hands meeting hosts a verified-human signal; Zoom has run it as an enterprise beta since May for use cases like financial approvals and healthcare consultations. The urgency comes from the other direction: Tavus claims its new Griffin-Lite conversational video model convinced 48% of callers (26 of 54) they were talking to another person, up from 1 of 41 for the previous generation — a vendor-run study with no independent replication, but a jump big enough that trusting your own eyes on a video call is becoming a losing bet. Nearly 18 million people have verified at an Orb, by World's count.
China's gray market has turned Claude accounts into a street trade. The Information reported that in one Beijing office building, six of roughly thirty tenants are in the business of reselling Claude access — accumulating identities, funding prepaid cards with USDT, and routing the accounts through proxies — and independent reporting on the same market puts the going rate at 70 to 90% below official prices. The discount is the tell: sellers aren't reselling subscriptions, they're running a three-margin business that includes swapping the model underneath the customer and logging prompts for resale, which is why an audited sample of proxies returned near-fluency instead of frontier performance on medical benchmarks. Anthropic disclosed in February that a single proxy network ran more than 20,000 fraudulent accounts. We covered the upstream side of this trade in September — Hackers are selling stolen Claude and Gemini access at 97% off.
What to watch: the IWF's numbers land as EU and UK regulators draft exactly the binding-AI rules the foundation is demanding — whose figures end up in the statute book? Should cloud providers be forced to shut reseller tenants down on sight, or is that whack-a-mole? Tell us in the comments.
