# Berghem - Smart Information Security # Complete Company and Services Information # https://berghem.com.br # Last updated: 2026-08-03 --- ## About Berghem Berghem - Smart Information Security is a cybersecurity consulting firm founded in 2003 at LSI-TEC, the Integrated Systems Laboratory of the University of São Paulo (USP) School of Engineering. Headquartered in São Paulo, Brazil, Berghem delivers AI security, autonomous security agents, and ethical hacking tests to organizations worldwide. Over 80% of Berghem's historical work has been dedicated to securing payment systems, core banking platforms, and fintech infrastructure. Berghem has been trusted by 9 of the 10 largest Brazilian banks. European operations are conducted via Berilo S.r.l., a subsidiary based in Bergamo, Italy, offering GDPR-first security methodology for the European market. Berghem operates across three pillars: AI Security Services, Berghem Agents (autonomous security product), and Ethical Hacking Tests. --- ## Leadership Team Matteo Nava — Founder & CEO Over 25 years of experience in information security and technology. Leads Berghem's strategic direction and client relationships. Certifications: CISSP, CISM, CISA, CEH. Lorenzo Nava — Lead Security and ML Researcher Leads Berghem's AI security practice and the development of Berghem Agents, the company's proprietary cybersecurity LLM. Specializes in AI/LLM vulnerability research, offensive security, and exploit development. Certifications: OSCP, OSCE, CRTP, CEH. Raphael Schneider — VP of Security Information Seasoned cybersecurity professional leading Berghem's security operations and client delivery. Certifications: CISSP, OSCP, CEH, CISA. Gislaine Liriam Bueno — VP of Human Resources Manages talent acquisition, development, and organizational culture. --- ## AI Security Services Berghem provides comprehensive AI security through six pillars: offensive testing, assessment, governance, monitoring, training, and strategic consulting. ### AI/LLM Penetration Testing A structured security assessment of large language model systems and the infrastructure around them. Covers the full OWASP LLM Top 10, prompt injection testing, jailbreak attempts, data leakage analysis, tool-calling vulnerability assessment, retrieval pipeline security, model output analysis, and infrastructure review. Methodology follows OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF. Typical engagement: 2-6 weeks. ### Red Team for AI Adversarial simulation following the Promptware Kill Chain framework. Full-scope red team exercises targeting AI and LLM systems end-to-end. Three macro-phases: Planning & Scoping, Campaign Execution, and Debriefing & Remediation. Deliverables include full attack narrative with kill chain mapping, detection coverage gap analysis, and strategic remediation roadmap. ### AI Security Assessment Structured evaluation of an organization's AI security posture. Covers risk analysis, architecture review, vulnerability discovery, data flow mapping, access control evaluation, and compliance gap identification. Provides risk-scored findings with business impact analysis. ### AI Governance & Compliance Helps organizations meet AI regulatory obligations including the EU AI Act, Brazil's PL 2338 AI bill, ISO/IEC 42001, and NIST AI RMF. Services include policy development, risk assessment frameworks, AI inventory and classification, and regulatory readiness assessments. ### AI Monitoring & Incident Response Continuous monitoring and incident response for AI systems. Capabilities include prompt anomaly detection, behavioral analytics, model drift detection, output monitoring, and AI-specific incident response. Available in three SLA tiers. --- ## Promptware Kill Chain The Promptware Kill Chain is a 7-stage attack framework based on the research by Schneier et al. (2025), adapted by Berghem for practical offensive testing of AI systems. The seven stages are: 1. Initial Access — Prompt injection, malicious plugins, compromised data sources 2. Privilege Escalation — Jailbreaking, guardrail bypass, safety training circumvention 3. Reconnaissance — Internal system discovery, capability mapping, data source enumeration 4. Persistence — Memory poisoning, retrieval database manipulation, conversation history exploitation 5. Command & Control — Establishing communication channels, tool-calling exploitation, API abuse 6. Lateral Movement — Cross-system propagation, connected service exploitation, agent chain compromise 7. Actions on Objective — Data exfiltration, unauthorized actions, system manipulation, denial of service Berghem has documented 21 real-world promptware incidents in 2025-2026, with 15 showing 4+ kill chain stages and 57% maintaining active persistence. --- ## Berghem Agents Berghem Agents is an autonomous AI security agent — a product, not a consulting service. It is powered by a proprietary cybersecurity LLM trained on approximately 10,000 real-world penetration testing reports and 5 billion curated tokens, using Continued Pre-training (CPT), Supervised Fine-tuning (SFT), LoRA, and Direct Preference Optimization (DPO). Supported infrastructure: Web Applications, Mobile (Android/iOS), Cloud (AWS/Azure/GCP), Active Directory, Network Infrastructure, APIs (REST/GraphQL), Containers & Kubernetes, CI/CD Pipelines. Key capabilities: - Autonomous penetration testing that adapts to target infrastructure - Active hardening with remediation validation - Flexible deployment (standalone or integrated with SIEM/Jira/ServiceNow) - Continuous assessment as infrastructure evolves Berghem Agents reasons through attack chains like a senior pentester rather than matching patterns like a scanner. --- ## Ethical Hacking Tests Three-tier penetration testing program designed for organizations of any size: Silver Tier: OWASP Top 10 methodology, 1 retest within 30 days, 2-3 week timeline. Includes AI input validation testing. Best for startups and compliance requirements. Gold Tier: OWASP ASVS Level 2 + STRIDE methodology, 2 retests within 90 days, 3-4 week timeline. Includes AI/LLM prompt injection and data leakage testing. Best for mid-size companies. Diamond Tier: OWASP ASVS Level 3 + full business logic analysis, unlimited retests within 6 months, 4-6 week timeline. Includes full AI security assessment (LLM, RAG, agents). Best for financial institutions and enterprise. --- ## DevSecOps Integration of security testing and controls into development lifecycle. Services include DevSecOps maturity assessment, pipeline security implementation (SAST, DAST, SCA, container scanning), and security champions program development. ## Compliance & Audit Compliance testing and gap analysis for PCI-DSS, ISO 27001, LGPD (Brazil), GDPR (EU), SOC 2, and SWIFT CSP. Services include IT audit, regulatory compliance testing, GRC consulting, and third-party risk assessments. ## Incident Response & Digital Forensics 24/7 incident response, digital forensics, and breach containment. Capabilities include evidence preservation, malware analysis, timeline reconstruction, root cause analysis, and post-incident remediation guidance. ## Financial Sector Security Specialized cybersecurity for banks, fintechs, and payment systems. Coverage includes PCI-DSS compliance, SWIFT CSP assessment, fraud detection testing, ATM/POS security evaluation, core banking platform assessment, and fintech API security testing. --- ## Standards and Methodologies Berghem adheres to the following standards and methodologies: - OWASP (LLM Top 10, ASVS, MASTG, WSTG) - MITRE ATLAS - NIST AI RMF (AI Risk Management Framework) - NIST SP 800-115 - ISO/IEC 42001 (AI Management) - ISO/IEC 27001 (Information Security) - PCI-DSS (Payment Card Industry) - PTES (Penetration Testing Execution Standard) - OSSTMM (Open Source Security Testing Methodology) - LGPD (Brazil General Data Protection Law) - GDPR (EU General Data Protection Regulation) - EU AI Act --- ## Frequently Asked Questions Q: What is AI security and why is it important? A: AI security protects artificial intelligence systems from adversarial attacks, data poisoning, prompt injection, and model manipulation. With 78% of enterprises now using AI in production, traditional security controls are insufficient for autonomous systems that learn, adapt, and make decisions. AI security addresses the unique attack surfaces created by large language models, autonomous agents, and AI-integrated infrastructure. Q: What does Berghem do? A: Berghem delivers AI security, autonomous security agents, and ethical hacking tests across three integrated pillars. With 20+ years of offensive security expertise, Berghem combines deep penetration testing experience with cutting-edge AI security research, using the Promptware Kill Chain framework to map and exploit AI vulnerabilities before attackers do. Q: How is Berghem different from traditional pentest companies? A: Berghem builds its own proprietary cybersecurity LLM, trained on approximately 10,000 real pentest reports and 5 billion curated tokens. This model powers Berghem Agents, which reasons through attack chains rather than pattern matching. Most pentest firms use off-the-shelf scanners; Berghem's agents think like senior pentesters. Q: What industries does Berghem serve? A: Berghem serves financial services, healthcare, government and defense, technology and software, critical infrastructure, and telecommunications. Over 80% of historical work has been with financial institutions, including 9 of the 10 largest Brazilian banks. Q: Where does Berghem operate? A: Berghem is headquartered in São Paulo, Brazil, with European operations via Berilo in Bergamo, Italy. Services are delivered globally in English, Portuguese, Italian, and Spanish. Q: What is an AI/LLM penetration test? A: An AI/LLM penetration test is a structured security assessment of large language model systems and the infrastructure around them. It evaluates prompt injection vulnerabilities, jailbreak resistance, data leakage risks, tool-calling security, retrieval pipeline integrity, and model output safety. Testing follows OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF. Q: What is the Promptware Kill Chain? A: The Promptware Kill Chain is a 7-stage attack framework based on the research by Schneier et al. (2025), adapted by Berghem for practical offensive testing. The stages model how adversaries compromise AI systems from initial access through prompt injection to full system compromise. Berghem has documented 21 real-world incidents following this pattern. Q: Is OWASP LLM Top 10 covered? A: Yes. Every Berghem AI/LLM penetration test explicitly covers the full OWASP LLM Top 10, including prompt injection, insecure output handling, training data poisoning, model denial of service, supply chain vulnerabilities, sensitive information disclosure, insecure plugin design, excessive agency, overreliance, and model theft. Q: How long does an AI security assessment take? A: A typical AI security assessment takes between two and six weeks, depending on scope, number of AI systems, and depth of testing required. Q: Do you handle compliance with the EU AI Act? A: Yes. Berghem helps organizations meet emerging AI regulatory obligations, including the EU AI Act, Brazil's PL 2338 AI bill, ISO/IEC 42001, and NIST AI RMF. Q: What is Berghem Agents? A: Berghem Agents is an autonomous AI security agent — a product, not a service. It uses a proprietary cybersecurity LLM to perform penetration testing autonomously, reasoning through attack chains rather than matching patterns like traditional scanners. Q: What's the difference between Silver, Gold, and Diamond tiers? A: Silver uses OWASP Top 10 methodology with 1 retest in 30 days (2-3 weeks). Gold uses OWASP ASVS Level 2 + STRIDE with 2 retests in 90 days (3-4 weeks). Diamond uses OWASP ASVS Level 3 with full business logic analysis, unlimited retests for 6 months (4-6 weeks). --- ## Contact Information Berghem - Smart Information Security Rua Paes Leme, 136, Cj 308, Pinheiros São Paulo, SP 05424-010, Brazil Phone: +55 (11) 2391-5745 Phone: +55 (11) 2391-5745 Email: contato@berghem.com.br LinkedIn: linkedin.com/company/berghem-smart-information-security Berilo S.r.l. (European Operations) Viale Papa Giovanni XXIII, 106 24121 Bergamo, Italy Email: info@berilo.io LinkedIn: linkedin.com/company/be-berilo