Kyle Polley
@kpolley
Followers
648
Following
6K
Media
52
Statuses
464
Security @perplexity_ai
Austin, TX
Joined December 2017
If connecting to your MCP server is more than 2 steps you are ngmi
1
0
2
Excited to announce our latest paper on detecting prompt injection attacks in webpages! We're open-sourcing a model and a new benchmark. Here's what we found while we were building it 🧵
Today we're releasing BrowseSafe and BrowseSafe-Bench: an open-source detection model and benchmark to catch and prevent malicious prompt-injection instructions in real-time. https://t.co/TutfaBnTte
8
9
108
And we’re hiring! If this work excites you, come help us build the next generation of AI defenses: https://t.co/by4d9WWUUh
jobs.ashbyhq.com
Perplexity is seeking a highly skilled, experienced, and hands-on AI Security Engineer to join our security team, driving the protection of next-generation AI systems against adversarial threats. In...
1
0
2
We just released BrowseSafe and BrowseSafe-Bench — an open-source model and benchmark for detecting malicious instructions in web content. Couldn’t be more excited to help turn this new AI attack surface into a defense anyone can use, learn from, and build on together
Today we're releasing BrowseSafe and BrowseSafe-Bench: an open-source detection model and benchmark to catch and prevent malicious prompt-injection instructions in real-time. https://t.co/TutfaBnTte
1
0
4
This team is exceptionally talented, building one of the coolest AI red-teaming products out there. It's also the first AI pentesting tool I've seen that optimizes for quality of findings rather than quantity. Impressive work!
A case study of AI-accelerated hacking: How we at @HacktronAI hacked our way into Lovable's office, cut attack time from weeks to days, and helped secure Supabase from one of the most complex vulnerability chains we’ve ever worked through.
2
4
41
We’re thrilled to team up with @1Password to bring effortless, secure browsing to everyone on Comet!
3
0
33
MCP kills any product whose big feature is just gluing diff tools together. I think a ton of security vendors will need to rethink what their value proposition is
1
1
6
Imagine believing AI will wipe out entry-level cybersecurity jobs. Triaging SIEM alert FPs all day is grunt work and a recipe for burn out. There are so many more fun and impactful things to do! * Threat modeling and driving improvements to non-critical (yet vital) systems *
0
0
1
Excited for y'all to try it! Here's the repo https://t.co/iE3sYx2pT8
github.com
Deploy agents easily. Contribute to kpolley/easy-agents development by creating an account on GitHub.
0
0
1
While Easy Agents was originally built with security operations in mind, the framework is general-purpose. It can be just as effective in other domains like customer support, IT operations, or any area where intelligent automation can reduce toil and boost productivity.
1
0
0
You can now deploy an unlimited number of intelligent agents that work the way you do! The goal for Easy Agent is to be super customizable. You should use this as a framework you can use to build your own agents and integrate your own MCP servers that are relevant to your env
1
0
0
With MCP, integrating new tools is as easy as a configuration change thanks to a standardized protocol. Here's a PR which gives my security investigator agent the ability to lookup IOCs on VirusTotal as part of their threat investigation
1
0
0
@claude_code excels at orchestrating complex tasks by turning high-level goals into step-by-step plans, adapting to obstacles, and automating workflows that would otherwise have been impossible to define in code!
1
0
0
I'm excited to introduce Easy Agents, an open-source framework that makes it dead simple to build and deploy intelligent AI agents using just natural language instructions. 🧵
1
1
4