Umanitek.ai's plan to save the Internet: erasing fake news, child abuse, and toxic content
An interview with Chris Rynning and Tomaz Levak, cofounders of Umanitek.ai, revealed how they want to immunize the web against deep-fakes, CSAM and algorithmic outrage.
Rynning and his long-time collaborator, Tomaz Levák, are the co-founders of Umanitek.ai, a Swiss startup racing to make artificial intelligence "privacy-first" and, above all, trustworthy. Their first product, the Umanitek Guardian Agent, acts like a decentralized sentinel: it fingerprints images or videos, queries a permissioned network, and blocks anything that’s been flagged as harmful, without forcing platforms to share the underlying files.
"We’re building a digital collective immune system," Levák said, describing how the agent asks rival platforms—YouTube, TikTok, Reddit—to verify whether a piece of content has been labeled illegal or non-consensual.
Fighting the good fight
Follow us on: 
Umanitek’s first partners read like opposing corners of the internet: the International Center for Missing & Exploited Children (ICMEC) and Aylo the holding company behind Pornhub, Brazzers and RealityKings. The logic is simple: ICMEC’s hashes of known child-abuse material and Alo’s database of verified adult content can both seed the decentralized knowledge graph that powers Guardian.
"If a law-enforcement agent uploads a confiscated hard drive, Guardian can tell in seconds which files are green-flagged and which 40 percent might be criminal,"... "That doesn’t exist today."
Chris Rynning, co-founder Umanitek.ai
Levák frames it as a public-private pact: "Companies won’t swap raw data—they value their datasets too much. But they will exchange signals if we remove the friction."
Beyond CSAM: deep-fakes, revenge porn, and AI hallucinations
The founders insist Guardian is only the opener. A consumer-facing tool is next, letting anyone protect their likeness from non-consensual deep-fakes. Levák wouldn’t commit to a date—"I wish it was yesterday," he laughed—but hinted at a cadence of "one product every few months."
Accuracy, they argue, comes from provenance, not bigger models. "Our agents always return verifiable proof: where the data came from, who owns it, whether it’s been tampered with,"... "That minimizes hallucinations because the guardrails are the data."
Tomaz Levak, cofunder Umanitek.ai
Middleware with a mission
Rynning sees Umanitek as a neutral infrastructure "model and territory-agnostic middleware" rather than another closed AI lab. The company is funded by a globetrotting cap table that includes Zoom and Canva’s first investor Bill Tai, The Social Dilemma creator Aza Raskin, and Antler founder Magnus Grimeland.
"We’re not motivated by exits," Rynning said. "It takes a village to clean up the internet. We need partners, not just capital."
More details in the full interview we had with Chris and Tomaz
You can also hear it in Spotify
