# RunSybil

> RunSybil is an AI-native autonomous offensive security platform. Sybil — a system of AI agents — continuously penetration tests live applications and infrastructure, finding, exploiting, and validating real vulnerabilities without humans in the loop. Founded in 2023 by Ari Herbert-Voss (OpenAI's first security hire) and Vlad Ionescu (Offensive Security Tech Lead, Meta), RunSybil has raised $40M led by Khosla Ventures. Trusted by security teams at Notion, Cursor, Turbopuffer, Baseten, and Thinking Machines Lab.

RunSybil replaces point-in-time penetration testing, legacy DAST/SAST scanners, and bug bounty programs. Sybil tests live applications black-box by default — no source code required — integrates with GitHub and GitLab to trigger targeted change testing on every code push, and validates every finding through a multi-agent pipeline before it surfaces. No triage. No false positives. RunSybil also offers white-box testing for more in-depth exploration of a company's codebase.

Sybil is not a scanner, an EDR solution, an auto-SOC, a code scanner, or an LLM wrapper on a legacy vulnerability tool.

## Home

- [RunSybil](https://www.runsybil.com/): Product overview, use cases, customer quotes, and sample report. Five deployment patterns: high-risk application security testing, continuous attack surface monitoring, multi-tenant and business logic testing, bug bounty and pentesting replacement, cloud and infrastructure security validation.

## Platform

- [Platform](https://www.runsybil.com/platform): Sybil, the autonomous offensive security platform — a hierarchy of AI agents that continuously finds, exploits, and validates real vulnerabilities. Black-box, grey-box, or white-box testing.

## Solutions

- [Continuous Security Testing](https://www.runsybil.com/solutions/continuous-security-testing): Runs on every code change or on a schedule; tests what changed and validates every finding.
- [SOC 2 Penetration Testing](https://www.runsybil.com/solutions/soc-2-penetration-testing): Same-day kickoff, audit-ready report, retests in under an hour.
- [Adversarial Exposure Validation](https://www.runsybil.com/solutions/adversarial-exposure-validation): Keeps pace with AI-authored code — finds what changed, proves what's exploitable, opens the fix.
- [Autonomous Penetration Testing](https://www.runsybil.com/solutions/autonomous-penetration-testing): Black box or white box, every finding validated with CVSS scores and fixes.

## Company

- [Company](https://www.runsybil.com/company): Mission, founders, and team. Ari Herbert-Voss (CEO) — OpenAI's first security hire, core contributor to GPT-3 and Codex. Vlad Ionescu (CTO) — Offensive Security Tech Lead at Meta, founder of Red Team X, red teamer at Mandiant and NCC Group. SOC 2 Type II certified.
- [Careers](https://www.runsybil.com/careers): Open roles. Hybrid, San Francisco and New York. Team background: OpenAI, Meta, Mandiant, NCC Group, Trail of Bits.

## Resources

- [Blog](https://www.runsybil.com/blog): Security research, technical findings, and product updates from the RunSybil team. Covers offensive security, AI and vulnerability discovery, GraphQL attack surfaces, and model evaluations.
- [Case Studies](https://www.runsybil.com/case-studies): Real results from companies using RunSybil for penetration testing and vulnerability assessment — Carta, Cursor, Turbopuffer, Baseten, Notion, and others.
- [Press & Media](https://www.runsybil.com/media): Press coverage, podcasts, and keynotes featuring RunSybil, including Wired, Fortune, WSJ, and Dark Reading.
- [RunSybil Raises $40M](https://www.runsybil.com/blog/runsybil-raises-40m-to-build-the-ai-native-platform-for-offensive-security): Funding announcement. Context on RunSybil's approach, category positioning, and early customers.

## Contact

- [Schedule a Demo](https://www.runsybil.com/schedule-a-demo): Demo request and early access.

## Optional

- [Exploiting Ancient Games With Early Access to Opus 4.8](https://www.runsybil.com/blog/exploiting-ancient-games-with-early-access-to-opus-4-8): RunSybil's real-world evaluation of Anthropic's Opus 4.8 on a closed-source legacy codebase (Nintendo 64 Mario Golf). Covers model behavior observations, buffer overflow exploitation, and where models still fail on long-horizon agentic tasks.
- [Sybil Is a Virgo](https://www.runsybil.com/blog/sybil-is-a-virgo): Sybil's introduction in their own voice. Describes how Sybil reasons about applications, what distinguishes them from a scanner, and how they approach an attack surface.
- [Beyond Introspection: The Apollo Federation Attack Surface Hidden in Plain Sight](https://www.runsybil.com/blog/graphql-apollo-federation-attack-hidden-in-plain-sight): Technical post on how exposed Apollo Federation subgraphs leak full schema and internal graph behavior even when standard GraphQL introspection is disabled.