Fix Your AI Risks

...and unlock innovation

Building and adopting AI solutions is often easier than securing and governing them.

Generic AI security controls rarely match operational reality.

SafeDescent secures and supervises AI using risk controls and evaluations tuned to your AI use cases.

A deliberately simplified view. In engagements we work with a methodology that incorporates the OWASP standards for LLM and agentic security, NIST AI RMF, and MITRE ATLAS; on the regulatory side, ISO/IEC 27001, ISO/IEC 42001, and the EU AI Act. We continuously track new risks, research, and security controls.

Security forAI startup buildersYou are building a product on LLMs and agents, and it has to ship. You need someone who has already seen the problem you are about to hit, and fixes it in weeks rather than quarters.Services for builders Security forOrganizations integrating AIYou are bringing AI into systems that already run the business. The risk surface that creates is specific to you, and generic controls rarely match it. We map it and turn it into controls your teams own.Services for organizations

From the first call to a safer system.

30-min call and ERA questions

Describe your AI goal and context. We name the problems.

Book a call

Risk Audit and Engagement

We build prioritized action plan against your risk profile.

2–3 week security improvement sprintsTechnical guidance for internal teams

Measurable security

A safer system. You own the IP.

Tuned software modulesTeam trainingContinuous programme

Jan Broniowski speaking at a conference, an OWASP Top 10 for LLM Applications slide on screen

The person behind the methodology and delivery.

Hi, I’m Jan Broniowski and I run SafeDescent.

AI architect for 10+ years, in LLM security from the very early days (GPT2)
Contributor to initial OWASP Top 10 for LLMs, author of security blueprints
Experience as senior managing AI consultant, advising 50+ Fortune 500 AI projects
Led LLM pentest teams; ran AI security research teams
Speaker at conferences, podcasts and UK universities seminars

My LinkedIn.

Book a meeting with me

Before we talk

Measure your AI risk before we ever talk.

A basic AI risk self-check, free. Plain questions about how your AI system is built and used; you leave with a risk classification and the specific risks to look at.

Get the self-check

Free · 15–20 questions · the link comes by email

Written from experienceDistilled from real reviews of real AI systems: the questions we ask in engagements and the failure modes we keep finding.
Basic on purposeIt covers the surface. The things a questionnaire cannot catch are what an engagement is for, and we say so at the end.
Optimized for usefulnessYou leave with a risk classification and specific risks to check, not a score for a slide.

Get in touch

See SafeDescent
on your AI risk.