6 min read
Prompt Injection, Explained Through Lakera's Gandalf
Lakera's Gandalf is the best on-ramp to prompt injection ever built. Here's the full defense taxonomy behind its levels, and why none of it is enough on its own.
Offensive security · AI · Web3 · Web2
We break your systems before attackers do.
DSEC Labs is a technical security firm delivering exploitation-driven audits across applications, infrastructure, smart contracts, and AI systems, with clear, actionable reporting your engineers can act on.
- AI
- LLM apps, agents & RAG
- Web3
- Smart contracts & DeFi
- Web2
- Apps, APIs, cloud & people
Research & write-ups
Latest from the blog
Field notes on offensive security across AI, Web3, and Web2.
7 min read
Build Your Own Gandalf: A Local Prompt-Injection Lab You Can Actually Break
Skip reading about how Gandalf-style guardrails work and build one yourself. A free, offline Python lab using Ollama that implements all eight defense patterns from this series, ready for you to attack.
11 min read
Beating Gandalf, Part 3: Defense in Depth, Level 7 & Gandalf the White
The final Gandalf levels stack every defense and then harden them with data from a million real attacks. They're beatable too, and that's the most important lesson in the whole series.
What we do
Security testing across your whole stack
From an AI agent to a multi-chain protocol or a web app, we tailor the engagement to your threat model.
AI Security
Adversarial testing for LLM apps, agents, and model pipelines.
2 services
Explore
Web3 Security
Smart-contract and full-protocol security across EVM and non-EVM chains.
2 services
Explore
Web2 Security
Offensive security across your apps, APIs, cloud, and people.
3 services
Explore
Advisory
Threat modeling, secrets reviews, and AI security training.
2 services
Explore
Why DSEC Labs
Depth over checklists. Impact over noise.
Automated scanners find the obvious. We find the vulnerabilities that actually get you breached: chained, exploited, and demonstrated, then explained in a way your team can fix and verify.
-
Manual & exploitation-led
Real attack paths, not scanner output dumped into a PDF.
-
Actionable reporting
Reproducible PoCs, severity ratings, and clear remediation.
-
Free fix verification
We re-test your fixes so you can prove the risk is closed.
-
Discreet & scoped
Tight scoping, NDAs, and careful handling of sensitive access.
Ready to find your real risk?
Tell us what you're building. We'll scope an engagement that fits your stack, timeline, and threat model.
Start a conversation