Perplexity launches AI security institute led by Purdue professor

Letting an AI agent browse the live web on your behalf is fantastic until a malicious webpage tricks it into exfiltrating your session cookies or ordering things you never asked for.

Perplexity is trying to get ahead of that nightmare. The company just announced the launch of the Secure Intelligence Institute, a dedicated research hub focused entirely on frontier AI security, privacy, and user trust.

To lead the effort, Perplexity tapped Dr. Ninghui Li as the institute's inaugural director. Li is a prominent security researcher, Samuel D. Conte Professor of Computer Science at Purdue University, and an ACM and IEEE Fellow who previously chaired the ACM CCS steering committee.

Consolidating agent security under one roof

This isn't Perplexity's first brush with AI security. Over the past year, the company has been steadily laying the groundwork alongside its product rollouts.

Back in April 2025, before publicly rolling out its Comet agentic browser, Perplexity brought in cybersecurity firm Trail of Bits to run a dedicated security audit and adversarial stress tests.

Comet launched three months later with a layered defense-in-depth setup. Then, in December 2025, Perplexity open-sourced BrowseSafe, a benchmark and detection model packing over 14,700 real-world attack scenarios across 14 categories of harm.

Earlier this month, the team also submitted a policy paper on securing autonomous agents to NIST. The new institute will bundle these efforts together, focusing on authentication, usable privacy, robust machine learning, and defenses built specifically for agentic AI.

Why it matters

As AI shifts from passive chatbots to active browsers that click buttons and manage workflows, security changes from an afterthought into a fatal vulnerability.

If you use tools like Comet, you are handing an AI direct access to your web sessions. A single successful prompt injection could compromise your data. Setting up an institute led by a respected academic shows Perplexity knows users and enterprises won't stick around if safety is handled like an afterthought.

In our view, this is a smart strategic play. Bringing in academic heavyweights and actively contributing to NIST guidelines gives Perplexity regulatory credibility that pure product startups usually lack.

The real challenge, though, is speed. Academic security research moves in semesters, while zero-day prompt injections move in hours. Whether SII can translate theoretical frameworks into real-time defenses inside Comet will determine if this is true security engineering or just an enterprise trust badge.