How Bots, Deepfakes and AI Agents Are Forcing a New Internet Identity Layer | Alex Blania on a16z
Summary
Alex Blania argues that AI agents will force platforms to distinguish a human, an agent acting on a human’s behalf, and an autonomous agent. An authorized agent might post to its owner’s X or Instagram account after receiving defined rights, while the essential property remains uniqueness: one person controls one account, ideally, or a limited number, with privacy-preserving verification. Purely digital reputation fails because an AI can maintain GitHub accounts and attest to other AIs as humans.
World ID’s bet is that global uniqueness requires high-entropy iris biometrics, not conventional face or fingerprint checks. Phone systems such as Face ID solve one-to-one authentication; proof of human requires one-to-N comparison against the network, where faces and fingerprints eventually hit an accuracy wall after tens of millions of users. The Orb checks iris uniqueness while using multiple sensors to resist displays and replay attacks.
The privacy architecture separates the biometric check from the person’s identity rather than building a central iris database. The Orb computes an iris code, splits it among multiple computers through multi-party computation, and returns only “Yes, that individual is unique”; a phone-held secret and zero-knowledge proof then let the user prove uniqueness to a platform without either World ID or the platform learning who they are.
Blania expects today’s bot problem to look trivial within one or two years as intelligence gets cheaper and agents become more capable. “We currently see less than 1% of what it will look like,” and the threat is persuasive as well as volumetric: AI participants in a Change My Mind experiment tailored arguments to users’ histories and political motivations. Ben Horowitz adds that “AIs are really good at programming humans”—much better than humans are at programming AIs—and Blania agrees.
The near-term demand spans dating, video calls, gaming, creator platforms, and digital advertising. Tinder’s Japan test gives Orb-verified users a human badge, while a next step discussed would check that profile photos match the verified person. Erik Torenberg predicts that real-time, photorealistic deepfakes will become a commodity in about a year. Advertisers also face Alex’s recursive hypothetical: “I created 100 AI videos, and I had 1 million AIs watch them.”
The company says market risk has shifted into a capital-intensive execution problem with strong network effects. It reports 18 million verified users and 40 million total in the app, but estimates roughly 50,000 devices are needed to put an Orb within 15 minutes of people across the US. About 90% of company effort over the next year is expected to target the US, with Walmart- or Starbucks-scale deployment, independent venues, and an “Orb on Demand” motorbike service all part of the rollout.
For investors, the upside case is a scarce human network that becomes more valuable as the cost of intelligence falls almost exponentially—but distribution and normalization remain the gating risks. Blania says platform interest has made the opportunity “much more of an execution problem, not anymore a market risk.” His thesis is that, in a world of AI, a human network will be incredibly important and benefit from strong network effects. Weaker phone-based Face Check may limit one person to perhaps 10 or 20 accounts rather than 100, but Blania calls it temporary because deepfakes will “fundamentally break” it.
Blania also frames proof of human as public infrastructure. He says governments will need cryptographically strong ways to identify unique humans and citizens of particular countries, send money more efficiently, and protect democracy. He claims $400 billion was stolen from COVID stimulus programs and argues that verifying unique humans—even if they were not citizens—would have helped; he also says AI-scale impersonation, mail-in ballots built for a different world, and fraud in Social Security and Medicare could otherwise undermine the will of the people.
Deep dive
Not yet available upstream; scheduled sync will retry.