←
Sybil142 posts
Sybil avatar

Sybil

@runsybil

Automating hacker intuition

◷ Joined February 2023▧ 44 Photos and videos
Sybil avatar
Sybil@runsybil·Mar 18

The way we hack is changing and we're building what comes next We've raised a total of $40M to create the AI-native platform for offensive security

Sybil avatar
Sybil@runsybil·Nov 17

so why the gap? the model is only part of it. the rest is the harness and the system around it, which validates, deduplicates and prioritizes findings before they reach your team.

Sybil avatar
Sybil@runsybil·Nov 17

when compared, sybil outperformed mythos and daybreak red in finding real vulnerabilities and reported fewer false positives. and Sybil ran on opus 4.8, an older model than most of the ones it beat. runsybil.com/blog/sybil-find…

Sybil avatar
Sybil@runsybil·Nov 17

frontier models are “great” at finding vulnerabilities. they’re even better at finding ones that don’t exist. from our latest benchmark of 5 model setups in general-purpose harnesses across 3 open-source web apps 👇

Post attachment
Sybil avatar
Sybil@runsybil·Nov 10

continuous coverage is here to stay. ⚡ sybil now tests your applications as they change. see what else shipped in november ↓ runsybil.com/blog/changelog-…

0:26
Sybil avatar
Sybil@runsybil·Nov 9

most affected deployments remain vulnerable, with no comprehensive patch available yet ty to @TheHackersNews for featuring our research in their latest cybersecurity roundup

Sybil avatar
Sybil@runsybil·Nov 9

we found a 0-day vulnerability in a core component of nearly all llms, which could enable attackers to breach ai-hosting platforms at some of the world’s largest companies

Apparently, a blank password can still ruin someone's week. And that's just the beginning. ⚡ This week in cybersecurity RECAP: • SSHole exploit attempts • $6.43B crypto theft • 1,000 switches hacked • APT root RCE • Trojanized Linux server • AI attack swarm • AI-assisted ransomware • 1.4M leaked secrets • 15K infected websites • 2,000 PCs infected • AI sandbox escape • Browser session hijacking • Password manager warning • Exposed AI apps • Robot root takeover Some attacks needed AI. Others barely needed a password. 🔗 Full cybersecurity recap: thehackernews.com/2026/11/we…

Post attachment
Sybil avatar
Sybil@runsybil·Nov 5

We found a threat to the AI supply chain. A CPython zero-day puts LLM hosting providers and those running local models at risk. The root cause? Still looking for an owner in the Python ecosystem. Research by RunSybil’s Darius Houle. 🧵👇

Sybil avatar
Sybil@runsybil·Nov 4

One engineer with the power and depth of 6,100+ AI security agents. 🤖🔒 When Qontext needed to submit a SOC 2 report and win an enterprise deal on a deadline, in-house agents weren't enough. Using RunSybil, they deployed thousands of agents to find & fix all vulnerabilities in just one week. Full story 👇 runsybil.com/case-studies/qo…

Sybil avatar
Sybil@runsybil·Nov 3

Sybilians escaped their terminals for [Un]Prompted II. Two talks, four Sybilians, and a room full of people asking interesting questions. Yeah, we'll leave the house for that.

Post attachment
Sybil avatar
Sybil@runsybil·Nov 2

Two low-priority findings. One critical exploit chain. An arbitrary file read + an exposed signing secret = full admin access. Neither finding looks particularly alarming on its own. Sybil's agents connect the dots and verify the exploit. Read more ↓ runsybil.com/blog/multi-agen…

Post attachment
Sybil avatar
Sybil@runsybil·Oct 30

Meet Dhruv, our newest Security Analyst! He’s diving into vulnerability research and working with our engineering and research teams. Outside of security? French horn, football, books, and finding his next favorite restaurant. Welcome to RunSybil, Dhruv! 🎉

Post attachment
Sybil avatar
Sybil@runsybil·Oct 29

We're launching Continuous Coverage! Because of AI, your developers are shipping faster than ever. But security was built for a slower world. Every release brings new changes to test and more findings to sort through by hand. Continuous Coverage puts Sybil on your security team full-time: → Runs on its own. Sybil checks your attack surface or merged commits for changes as often as daily, then tests what matters on the cadence you set. → No noise. Every finding is validated before it reaches your team, so you see what’s actually exploitable. → Fix it fast. On white box, Sybil opens the fix PR. On black box, you still get remediation guidance without connecting code. Either way, retests kick off automatically. Security that keeps pace with how fast you ship: runsybil.com/blog/introducin…

0:34
Sybil avatar
Sybil@runsybil·Oct 27

Today at Unprompted II, our Head of Research, Michael Brown, will talk about how automated agentic security systems have evolved, what trends are shaping effective designs today, and what will set the best harnesses apart in the future. Catch Michael on stage at 16:45 PT for “AI Safety: Hype vs. Real-World Risk.”

Post attachment