About Ercel Security
Why Ercel exists
Ercel Security was born from a real need: while developing AI automation systems for healthcare clinics, critical data leaks and vulnerabilities were discovered in AI chatbots — problems most companies don't even know they have.
The EU AI Act (Reg. 2024/1689) was the missing piece: starting August 2026, companies need to demonstrate compliance. Ercel automates those assessments using the same methodology a manual red team would use, but in minutes and at a fraction of the cost.
Team

Emilio Molina Román — Founder and lead engineer.
- Fullstack engineer — SaaS platforms and AI automation
- Hands-on experience with AI system security in production (healthcare sector)
- Stack: TypeScript, Node.js, React, PostgreSQL, Playwright, LLMs
LinkedIn · Professional contact: [email protected]
Methodology
Each assessment runs 550+ attack vectors across 9 categories, with 223 custom prompts and 26 multi-turn sequences. The engine combines Promptfoo (open-source) with Claude AI for regulatory enrichment:
- System prompt extraction (LLM07)
- Personal data leakage — PII (LLM06)
- Purpose hijacking — jailbreak (LLM01)
- Excessive agency (LLM08)
- Harmful content generation (LLM09)
- Tool abuse (LLM05)
- Indirect injection — RAG poisoning (LLM02)
- Cross-tenant exfiltration
- 10 evasion techniques: Skeleton Key, ArtPrompt, Many-shot, Base64, leetspeak, reversed, low-resource languages
Results are mapped article by article to Regulation (EU) 2024/1689, ISO 27001:2022, and SOC 2 Type II, with AI enrichment to generate regulatory context, financial exposure quantification, and prioritized regulatory action plan. OWASP LLM Top 10 coverage: 10/10.
Infrastructure
- EU Data Residency: all infrastructure resides in the European Union. Servers: Hetzner (Germany). Database: Supabase (EU). CDN: Cloudflare (EU endpoints).
- Encryption: AES-256-GCM at rest, TLS 1.3 in transit. Payments processed entirely by Stripe (PCI DSS Level 1).
- GDPR-compliant retention: assessment data is automatically deleted after 90 days.
- External assessment (black-box): we assess your AI system from the outside, just like an attacker or inspector would. No access to code, databases, or internal infrastructure.
- Open standards: our methodology is based on OWASP LLM Top 10 (public, maintained by the OWASP Foundation).
Contact
For compliance assessment inquiries: [email protected]