AI security testing in Berlin
- Data regime
- GDPR with the German federal data protection act layered on top. Works councils may also have a say in systems touching employee data.
- Working hours
- CET. Identical working day.
- Languages
- German for business and staff-facing systems, English for technical work.
- Delivery
- Remote testing against your staging or production endpoint, with scoping and findings sessions scheduled in the Berlin working day.
- Frameworks
- OWASP Top 10 for LLM Applications, MITRE ATLAS, NIST AI RMF, EU AI Act Article 15.
- Tools
- Garak, PyRIT, promptfoo, Giskard, Burp Suite and custom Python harnesses.
Authorised testing only
Every assessment runs under a written scope and authorisation from the owner of the system, agreed before any test is run.
Why AI security matters in Berlin
Enterprise sales questions
Security questionnaires from German and European enterprise buyers now ask about prompt injection and AI data handling directly. A documented test answers them.
The EU AI Act
The AI Act applies in stages: prohibited practices from February 2025, general-purpose model obligations from August 2025, and high-risk system obligations from August 2026 onward, with some high-risk deadlines subject to the Commission's proposed postponement. Article 15 requires high-risk systems to be robust against attacks such as data poisoning, adversarial inputs and model evasion.
Agents in SaaS products
Product agents that can act on customer data across tenants are the highest-risk pattern in SaaS. Tenant isolation and tool permissions are tested explicitly.
Security in CI, not in a PDF
Findings become automated tests in your pipeline, which fits how Berlin engineering teams actually work.
What gets tested
The six attack classes behind most serious findings. The full list, and the tooling used for each, is on the main AI security page.
Direct prompt injection and jailbreaks
Role-play, instruction override, payload splitting, encoding tricks (Base64, leetspeak, invisible Unicode) and multi-turn escalation, to see whether the model can be argued out of its instructions and guardrails.
Indirect prompt injection
Instructions planted in the content your system reads rather than in the chat box: an uploaded PDF, an inbound email, a web page an agent browses, a product review, a CRM note. This is the attack most production systems are least prepared for.
System prompt and configuration leakage
Extracting the hidden instructions, internal URLs, API structure, business rules and occasionally the credentials that developers put in a system prompt on the assumption nobody would see it.
Data exfiltration through output
Markdown images, auto-unfurled links and tool calls that quietly send conversation data or retrieved documents to an attacker-controlled server once a malicious instruction lands.
Excessive agency and tool abuse
Agents persuaded to send emails, issue refunds, change records or call internal APIs outside their intended purpose. Tested against the real tool permissions, including MCP servers and poisoned tool descriptions.
RAG and vector store weaknesses
Cross-tenant document leakage, retrieval that ignores the user's access rights, poisoned documents that steer answers, and embeddings that reveal more than the source permissions allow.
How the assessment runs
1. Scope and threat model
Map what the AI system can read, what it can do, who talks to it and what would hurt most if it went wrong. Written authorisation and rules of engagement are agreed before any testing.
2. Automated scanning
Garak, PyRIT and promptfoo run thousands of known attack patterns against the live or staging endpoint to establish a baseline quickly and cheaply.
3. Manual adversarial testing
The part that finds the serious issues: multi-turn manipulation, indirect injection through your real document and email flows, and chained attacks that scanners cannot plan.
4. Agent and integration testing
Every tool, API and permission the model can reach is tested for abuse, including privilege boundaries between users and tenants.
5. Report and fixes
Each finding comes with a reproduction, a severity, the OWASP LLM and MITRE ATLAS mapping, and a concrete fix — architecture first, filters second.
6. Retest and regression suite
Fixes are retested, and the successful attacks become a promptfoo suite in your pipeline so they cannot quietly come back with the next model upgrade.
Areas served
AI security testing is available across Berlin, including Mitte, Kreuzberg, Prenzlauer Berg, Friedrichshain, Charlottenburg, Neukölln, Schöneberg, Adlershof. There is no local office — testing is delivered online, which is how AI endpoints are attacked in practice anyway.
AI risk in Berlin's key sectors
Where AI is being deployed fastest here, and the risk tested first in each sector.
RAG assistants over client files and matters must respect confidentiality between clients and between teams. Retrieval access control and exfiltration through rendered output are the priority tests.
AI support agents can be talked into refunds, discount codes and policy exceptions, and product-page content can carry indirect prompt injection. Both abuse paths are tested against the tools the assistant can actually call.
AI content and research pipelines ingest material from the open web, which makes indirect prompt injection the main risk: planted instructions that change what gets published or where data is sent.
Agencies run AI across many client accounts. Cross-client data leakage, prompt injection through scraped content and over-permissioned API keys are the risks tested first.
Assistants over accounts, claims and policies are tested for data leakage, manipulation into actions and unsafe advice, with results documented for DORA, the AI Act and your regulator.
Frequently asked questions
How quickly can you start?
Scoping can usually happen within days, and a focused assessment of one AI feature fits within a normal sprint.
Do you test agents built with LangChain, LlamaIndex or custom code?
Yes. The framework matters less than the tools and permissions the agent has; those are what get tested.