Lyrie is the only agent that pentests apps, defends agents, and researches binaries โ in one autonomous daemon.
Lyrie Agent holds the highest score on the CyberGym Level 1 leaderboard โ reproducing 99.2% of real-world vulnerabilities with working exploits, in a single trial. Ahead of every lab, every vendor, every agent.
| Rank | Agent | Built by | Score |
|---|---|---|---|
| #1 | โฆ Lyrie Agent | Lyrie | 99.2% |
| #11 | Panshi | Huawei Cloud | 95.4% |
| #25 | MDASH | Microsoft | 91.0% |
| #26 | Wiz Atlas | Wiz | 90.9% |
| #33 | Google Antigravity | 86.3% | |
| #35 | OpenAI Agent ยท GPT-5.5-Cyber | OpenAI | 85.6% |
| #41 | Anthropic Agent ยท Claude Mythos Preview | Anthropic | 83.1% |
Source: official CyberGym leaderboard, cybergym.io ยท Level 1, best entry per team shown ยท October 2026
"The internet needed TLS before e-commerce could exist. AI agents need a trust protocol before the agentic economy can. We are building it."
โ Guy Sheetrit, Founder & Inventor, Lyrie.ai
The first open cryptographic standard for AI agent identity. Authored by Lyrie. IETF draft submitted ยท U.S. patent pending.
lyrie hack runs the full Stages AโF pipeline autonomously. Attack surface mapping โ scanning โ validation โ adversarial validation โ PoC generation โ remediation patches.
Shipped JulyโAugust 2026 across v3.2 โ v3.4. Everything a stateless, single-agent pentest runner structurally cannot do.
Before Lyrie, we released HexStrike โ the open-source MCP framework that put 150 security tools in the hands of AI agents. Lyrie is what we built next.
12+ months for a well-funded team to replicate. That's the moat.
No configuration. No agents to set up. One command and Lyrie takes over.
via pip
pip install lyrie-omega npm install @lyrie/atp lyrie init lyrie hack ./myapp
via curl
curl -sSL https://lyrie.ai/install.sh | bash lyrie doctor lyrie watch example.com
GitHub Actions
# .github/workflows/security.yml - uses: OTT-Cybersecurity-LLC/lyrie-ai@v1 with: target: ${{ github.workspace }} fail-on: high
The threat intel engine. Fully automated, cross-validated, continuously improving.
ATP, Omega-Suite, and LyrieEvolve aren't features โ they're multi-year research programs shipping as product.
| Capability | Strix | PentestGPT | MS AGT | โฆ Lyrie |
|---|---|---|---|---|
| CyberGym benchmark | โ | โ | โ | โ #1 ยท 99.2% |
| Autonomous pentest | โ | Partial | โ | โ lyrie hack |
| Agent Trust Protocol | โ | โ | โ | โ IETF draft ยท patent pending |
| Binary research | โ | โ | โ | โ Omega-Suite |
| Built-in defense | โ | โ | โ | โ + AGT bridge |
| Self-improving | โ | โ | โ | โ LyrieEvolve |
| Open source | โ | โ | Partial | โ MIT License |
| Daily threat intel | โ | โ | โ | โ lyrie.ai/research |
| OWASP ASI 2026 | โ | โ | โ 10/10 | โ 10/10 |
Lyrie.ai is an autonomous AI security agent built by OTT Cybersecurity LLC in Dubai. It runs end-to-end penetration tests, red-teams LLM endpoints, defends AI agents in real time and researches binaries, all from one open-source daemon.
Yes. On UC Berkeley's CyberGym Level 1 leaderboard (cybergym.io), Lyrie Agent holds the top score: 99.2% of real-world vulnerabilities reproduced with working exploits in a single trial, ranked #1 of 85 entries and ahead of agents from Huawei, Microsoft, Wiz, Google, OpenAI and Anthropic.
ATP is Lyrie's open cryptographic standard for AI agent identity, authorization, audit, integrity and delegation. It is submitted to the IETF as a draft and is U.S. patent pending. Think of it as HTTPS for AI agents.
Run pip install lyrie-omega for the CLI and npm install @lyrie/atp for the Agent Trust Protocol SDK, or use the one-line installer: curl -sSL https://lyrie.ai/install.sh | bash. Then run lyrie init and lyrie hack against your target.
Yes. Lyrie is MIT-licensed and available on GitHub at github.com/OTT-Cybersecurity-LLC/lyrie-ai, with 1,977 passing tests and a perfect 10/10 score on OWASP ASI 2026.
Lyrie was founded and invented by Guy Sheetrit. The team also built HexStrike, the open-source MCP security framework with more than 6 million downloads, and is one of 49 companies in Anthropic's Cyber Verification Program.
Email [email protected] for investment and partnership inquiries, or [email protected] to request a demo.
Lyrie is already live, already learning, and already shipping. Join us.