Elloe AI wants to be the ‘immune system’ for AI — check it out at Disrupt 2025

5 months ago 144
Image Credits:Elloe AI

11:20 AM PDT · October 28, 2025

Owen Sakawa, the laminitis of Elloe AI, wants his level to beryllium the “immune strategy for AI” and the “antivirus for immoderate AI agent.”  

The idea, Sakawa said successful an interrogation days earlier the TechCrunch Disrupt conference, wherever Elloe AI is simply a Top 20 finalist successful the Startup Battlefield competition, is to adhd a furniture to companies’ LLMs that checks for bias, hallucinations, errors, compliance issues, misinformation, and unsafe outputs.  

“AI is evolving astatine a precise accelerated pace, and it’s moving this accelerated without defender rails, without information nets, without mechanics to forestall it from ever going disconnected the rails,” Sakawa (pictured above) said. 

Elloe AI is an API oregon an SDK, a module that sits connected apical of an AI model’s output layer, an “infrastructure connected apical of your LLM pipeline,” arsenic Sakawa explained. “And it sits determination fundamentally fact-checking each azygous response.” 

The startup’s strategy has its ain layers, oregon “anchors,” arsenic Sakawa enactment it.  

The archetypal anchor fact-checks the LLMs effect against verifiable sources. Then, the 2nd anchor checks if the output violates immoderate regulations, specified arsenic the U.S. wellness privateness instrumentality HIPAA, the European far-reaching information extortion and privateness instrumentality GDPR, oregon if it exposes immoderate Personal Private Information (PII). The past anchor is an audit way that shows however each the erstwhile decisions were made and allows regulators oregon anyone auditing the strategy “to analyse the bid of thought for that exemplary from wherever it made the determination the root of that decision, the assurance people of each those decisions,” according to Sakawa.  

To beryllium clear, Sakawa said Elloe AI is not built connected an LLM, due to the fact that successful his opinion, having LLMs checking different LLMs is conscionable putting a “Band-Aid into different wound.” Elloe AI’s strategy does usage AI techniques, though, specified arsenic instrumentality learning. And determination are humans successful the loop: Elloe AI’s employees, who support up with caller regulations connected information extortion and idiosyncratic protection, Sakawa said.  

Techcrunch event

San Francisco | October 27-29, 2025

If you privation to larn much astir Elloe AI — portion besides checking retired dozens of different companies, proceeding their pitches, and listening to impermanent speakers connected 4 antithetic stages — articulation america astatine Disrupt, October 27 to 29, successful San Francisco. Learn much here.   

TechCrunch Disrupt 2025 nary  anniversary

Lorenzo Franceschi-Bicchierai is simply a Senior Writer astatine TechCrunch, wherever helium covers hacking, cybersecurity, surveillance, and privacy.

You tin interaction oregon verify outreach from Lorenzo by emailing lorenzo@techcrunch.com, via encrypted connection astatine +1 917 257 1382 connected Signal, and @lorenzofb connected Keybase/Telegram.

Read Entire Article