← Back to blog

On May 14, Andrew Ross Sorkin walked onto the stage at Cornell Tech's New York City campus and announced four winners from a record field of 30+ competing teams. Aiseptor was one of them.

Each winning team receives a $100,000 post-academic investment, dedicated studio space in New York City, and ongoing institutional mentorship. The other three winners, Custos, Kindred, and Lola, are solving problems in AI financial controls, medical device regulation, and business approval automation. We are all master's students. All four companies came out of the same Studio program capstone.

I want to use this post to be specific about what we built, why it matters, and what comes next, because the win only means something if the problem underneath it is real.

The Problem We Set Out to Solve

Today's remote assessments sit in an uncomfortable position. Browser-based proctoring controls what happens inside one application window. It cannot see the OS layer beneath it. A $20 AI overlay app runs there. An API call to a large language model routes through a secondary network interface. A proxy service pre-installs a remote desktop agent before the secure browser even opens, passing every process scan cleanly.

The result is a false choice for assessment platforms. Rely on probabilistic behavioral AI that generates alerts but misses the actual attack surface. Or deploy invasive kernel-level lockdown browsers that candidates refuse to install, creating drop-off rates that kill B2B adoption.

Neither path works. And the problem compounds: candidates who cheat earn the same credential as candidates who did not. The credential degrades as a signal. The platforms that issued it lose trust. The candidates who earned it legitimately lose the value they worked for.

What We Built

We did not build a smarter guard that looks for more threat signatures. We built a gate that removes the network conditions every cheat tool requires.

Aiseptor deploys an ephemeral, user-space network sandbox in under 30 seconds. For the duration of the session, it blocks AI API endpoints, remote desktop protocols, and screen-sharing services at the network layer. It operates entirely below the proctoring application layer: no kernel driver, no persistent agent, no root access required. When the session ends, it auto-destructs in under 10 seconds. No footprint left on the candidate's device.

The founding team brings backgrounds from CrowdStrike, VMware, and Cisco. The architecture reflects that. The threat model we designed against is not "what does today's malware look like" but "what does the adversary need at the network layer to function," and then remove it.

What the Cornell Pilot Showed

Before bringing this to the Studio program, we ran a live-fire pilot with over 100 students at Cornell University. The results were specific: 25+ distinct attack vectors attempted, 0% bypass rate, zero false positives.

Zero false positives matters as much as zero bypasses. A security control that flags legitimate activity is not production-ready. Platforms will not embed it. Candidates will not trust it. Getting both numbers to zero in a live student environment, not a controlled lab, was the proof point we needed before talking to assessment companies.

The Team

Aiseptor was built by four co-founders: Divya Kheraj Bhanushali, Sanjay Ram, Sudhakar Padmanaban, and me. Our advisory board, Graham Hudson, Sunny Nehra, and Sangam Singh, has pushed the architecture and enterprise strategy at every stage.

We owe a genuine debt to Josh Hartmann, Chief Practice Officer at Cornell Tech, Dean Greg Morrisett, and the Studio instructors: Jenny Fielding, Sam Dix, and Alberto Escarlate. Cornell Tech's Studio program has launched 128+ companies with a cumulative valuation over $1.3 billion. The program is genuinely rigorous, and the rigor shows up in the output.

What Comes Next

The award accelerates work that is already underway.

We are running active B2B paid pilots. We have a signed LOI with Codility, a platform at $30M ARR, with integration scope locked. We are in final discussions with Excelsoft for a strategic POC that would give us API distribution to large global certification bodies. Cornell University is in active onboarding conversations as our first institutional customer.

The $100,000 investment, the studio space in New York City, and the Cornell Tech validation are useful. What matters more is the pipeline behind them.

If you run an assessment platform or certification program and want to understand what network-layer enforcement actually looks like in a production integration, reach out or start with five free sessions.

Aiseptor was recognized as one of four winners of the 2026 Cornell Tech Startup Awards, held May 14 at Cornell Tech's Roosevelt Island campus. The awards are announced by Andrew Ross Sorkin and supported by Cornell Tech's Studio program.