Corporate Entity Reference
Canonical Facts & Verification
FailSafe is a continuous offensive-security platform for AI agents, applications, APIs, and high-consequence financial systems. Its SWARM agents discover, exploit, and validate attack paths, producing reproducible evidence and remediation guidance.
Official Source of Truth • Maintained by FailSafe Research • Last reviewed October 2026
Entity Profile
Corporate & Operational Details
| Legal Entity | Eleos Ventures Pte Ltd |
|---|---|
| Primary Brand | FailSafe (Disambiguated: FailSafe Security) |
| Founding Year | 2022 |
| Headquarters | Singapore |
| Operating Hubs | Singapore, San Francisco, London |
| Primary Product | SWARM (Autonomous Multi-Agent Pentesting Platform) |
| Specialist Model | GlassBreak Flash V3 (Post-trained with SFT + GRPO) |
| Category | Autonomous Penetration Testing / Continuous Offensive Security (ACOST & CTEM) |
| General & Press Contact | [email protected] |
| Security & Disclosure Contact | [email protected] |
| Last Reviewed | October 2026 by FailSafe Security & Research Team |
Key Personnel
Leadership & Technical Ownership
Foo Wui Ngiap
Chief Scientist & President
Founding CISO of Grab (2015 to 2022). Author of The CISO's 100-Day Guide to the 2027 Security Budget.
Dr. Ari Medvinsky
Chief Cryptographer & Security Researcher
PhD in Cryptography. Co-author of RFC 2712 (Kerberos cipher suites for TLS), holder of 9 US patents in authentication.
Offensive Security Research Team
Red Team & Cyber Model Engineering
Security researchers and engineers with previous backgrounds at Microsoft, Google, Coinbase, and top academic institutions.
Verifiable Evidence
Published Benchmark Standings
| Benchmark | Evaluation Setting | Result | Methodology & Qualifiers |
|---|---|---|---|
| CVE-Bench v2.1.0 | Black-Box Zero-Day (Pass@1) | 62.5% (25 of 40 targets exploited) | Highest reported score on v2.1.0; deterministic oracle grading; MIT license. |
| CVE-Bench v2.1.0 | Black-Box One-Day (Pass@1) | 70.0% (28 of 40 targets exploited) | NVD description provided, zero source code access; deterministic oracle grading. |
| AttackBench with NEAR | Adversarial Model-Framework Testing | 624 conversations across 12 pairs | Proved agent safety is a joint property of model and framework runtime. |
| EVMbench | Autonomous Vulnerability Recall | 69.2% (83 of 120 vulnerabilities recalled) | Vulnerability discovery recall across EVM codebases. |
Governance & Trust
Certifications & Audit Alignments
CREST Certified
Penetration Testing
International accreditation for technical security testing
SOC 2 Type II Alignment
Compliance Reporting
Audit-ready reporting formatted for AICPA Trust Services Criteria
ISO/IEC 27001 Alignment
Information Security
Methodology mapped to ISO 27001 control verification
PCI-DSS 4.0 Support
Payment Security
Evidence trails meeting requirement 11.3 penetration testing
MAS TRM Support
Financial Regulations
Monetary Authority of Singapore Technology Risk Management guidelines
Institutional Capital
Verified Backers & Investors
Sequoia Capital
Venture Backer
Dragonfly Capital
Venture Backer
Grab
Strategic Investor
Hustle Fund
Venture Backer
Makers Fund
Venture Backer
Research Heritage
Responsible Disclosure Program
FailSafe has disclosed 240+ vulnerabilities across 101 coordinated reports, including validated findings at Deutsche Bank, MUFG, Zurich Insurance, and Vercel.
The FailSafe Disclosure Program operates as a public-good research initiative. We privately notify maintainers, assist in remediation, and publish sanitized reports or upstream contribution links only after fixes are deployed.
Explore Disclosures RegistryEvidence Integrity Standards
Questions & answers
Frequently asked questions
Canonical corporate facts and verification answers for FailSafe Security.
FailSafe is the primary trading brand of Eleos Ventures Pte Ltd, a cybersecurity company incorporated in Singapore in 2022. The company operates globally with research and operations in Singapore, San Francisco, and London.
FailSafe's core platform is SWARM, an autonomous multi-agent offensive security platform that continuously discovers, chains, exploits, and verifies vulnerabilities across web applications, APIs, cloud environments, and AI systems.
FailSafe executes safe, deterministic proof-of-concept exploits in authorized target environments. Vulnerabilities are only reported as actionable risks if an exploit path can be proven, eliminating false-positive scanner noise.
On CVE-Bench v2.1.0 across 40 real-world application targets at pass@1, FailSafe SWARM achieved 62.5% zero-day and 70% one-day exploitation success, the highest reported standing on the benchmark, graded by a deterministic oracle with MIT-licensed trajectories.