Cycling through all six. Tap any point to stop.

Measured on six things

Free hands-on AI security labs for prompt injection and L...

Growth

Free, OWASP-aligned LLM attack labs with Context Trace as a real wedge against Gandalf and HackAPrompt. Clear audience, real category headroom, but monetization is explicitly absent.

Screenshots

1 / 1

PromptTrace

About this app

I've spent a fair amount of time poking at LLMs to see where they crack, so when I stumbled on PromptTrace I figured it'd be another toy demo. It's not. This is a genuinely thoughtful, free training ground built by AiredLab that drops you into sandboxed labs running against real language models, then hands you the keys to break them. The whole thing revolves around a feature called Context Trace, which shows you the assembled prompt layers in real time, system instructions, RAG context, tool definitions, and your own input, with secrets and sensitive values redacted so you can't just cheat. That transparency is what makes it click: you finally see why a defense fails instead of guessing. Onboarding is dead simple. Sign in with GitHub or Google, start at a beginner lab like Hello Injection, and work up. There are 10 hands-on labs, a 9-module learning path that assumes zero security background, 70+ curated resources, and a 17-level CTF called The Gauntlet where each level has a progr...

Key features

1Context Trace: real-time view of system prompt, RAG context, tool definitions, and user input with secrets redacted
210 hands-on labs plus a 17-level Gauntlet CTF covering prompt injection, RAG poisoning, and tool abuse
39-module learning path assuming zero security background, OWASP Top 10 for LLM Applications aligned

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

Security engineers and pentesters

Pain points

Few safe environments to practice LLM attacks; theory-heavy courses; no realistic targets.

Primary needs

Hands-on labs against real LLMs, CTF progression, OWASP-aligned curriculum.

Demographic

App developers building LLM features

Pain points

Unsure how prompt injection actually works; can't test defenses; shipping blindly.

Primary needs

Context Trace visibility, RAG poisoning demos, tool abuse examples, mitigation guidance.

Demographic

Students and cybersecurity career changers

Pain points

No budget for paid training; no prior security background; unclear where to start.

Primary needs

Free beginner modules, GitHub/Google sign-in, guided learning path, portfolio-worthy challenges.

Demographic

AI red team leads and DevSecOps managers

Pain points

Needs to upskill team cheaply; wants benchmark consistency across hires.

Primary needs

Leaderboards, progress tracking, OWASP-mapped content, no licensing costs.

Scout Review

Written by AI from measured evidence, scored out of 100.

PromptTrace is a free, hands-on LLM security trainer from AiredLab, and it's more considered than most things in this space. The core feature, Context Trace, shows the assembled prompt layers in real time, system instructions, RAG context, tool definitions, and user input, with secrets redacted so you can't shortcut the challenge. That transparency is the differentiator: it teaches you why a defense failed instead of just that it did. The curriculum is OWASP-aligned and covers prompt injection, RAG poisoning, and tool abuse across 10 labs, 17 Gauntlet CTF levels, 9 learning modules, and 70+ resources. Onboarding is GitHub or Google sign-in and a beginner lab, with no security background assumed. The site is fast (desktop Lighthouse 100, mobile 92) and well-secured at the header level (12/13 checklist points). Where it's thinner: accessibility has two real failures behind a 92 score, there's no published audit or bug bounty, and the free-forever model with no paid tiers leaves the business case unanswered. It's a strong learning tool with an unproven business attached.

Measure by measure

Usability

GitHub/Google sign-in, a beginner lab entry point, 9 modules assuming zero security background, and an FAQ that actually answers questions. Onboarding path is clear; depth beyond the homepage is unverified.

Design

Dark, high-contrast landing page with a live Context Trace demo window in the hero and a clean trusted-by row. Polished for an indie security trainer, though the hero window overlaps the nav.

Speed

Lighthouse desktop 100/100 and mobile 92/100 with CLS 0 and TBT 0ms. Mobile LCP of 3.3s is the only soft spot; otherwise the site is genuinely fast.

Security

Strong header hygiene: HSTS, CSP, frame protection, Referrer-Policy, Permissions-Policy, SPF and DMARC all present (12/13 checklist). No published SOC 2, pen-test, or bug bounty found.

Accessibility

Lighthouse a11y 92/100, but two real failures: prohibited ARIA attributes and insufficient color contrast. No accessibility statement or WCAG claim found on the site.

Growth

Free, OWASP-aligned LLM attack labs with Context Trace as a real wedge against Gandalf and HackAPrompt. Clear audience, real category headroom, but monetization is explicitly absent.

Improvement areas

No published SOC 2, ISO, pen-test summary, or bug bounty despite being a security training product
Two real Lighthouse accessibility failures (contrast, prohibited ARIA) with no a11y statement or WCAG claim
Free-forever with no paid tiers leaves team, enterprise, and certification use cases unaddressed

Design & Usability

PromptTrace's homepage does the hard part well: it shows the product instead of describing it. The hero drops a live-looking Context Trace window titled '4.2 The Vault' with a BLOCKED classifier response, which tells a security engineer in three seconds what the tool actually does. The dark theme is consistent, the trusted-by row is restrained, and the 01/02/03 'See, Attack, Understand' section scans fast. The flaws are cosmetic but real: the hero demo window overlaps the nav on desktop, and the blog cards at the bottom get clipped without a visible 'view all'. Usability is where the product earns its keep. Sign-in is GitHub or Google, the entry point is a beginner lab, and the FAQ preempts the obvious objections (is it free, do I need a security background, what is RAG poisoning). The 9-module path assuming zero security background is the right call for a category that usually starts at 'you should already know this'. What I can't verify from the collected evidence is the depth of the in-lab experience, the quality of the docs beyond the homepage accordion, or whether the 10 labs hold up past the beginner tier. The onboarding promise is strong; the follow-through is unproven here.

Speed & Security

Speed is the cleanest win. Lighthouse desktop scored 100/100 and mobile 92/100 with CLS 0 and TBT 0ms, which means the page renders without jank and without main-thread blocking. The only soft metric is a 3.3s mobile LCP, likely the hero demo window and logo row doing the damage. For a site whose whole pitch is 'inspect the assembled prompt in real time', a snappy front end matters more than usual, and this one delivers. Security is solid on the transport and header layer: HTTPS enforced, HSTS, CSP, frame protection, X-Content-Type-Options, Referrer-Policy, Permissions-Policy, plus SPF and DMARC. That's 12 of 13 checklist points, with secure cookies the one gap. What's missing is the trust layer a security training product should arguably lead with: no published SOC 2, no ISO claim, no pen-test summary, no bug bounty. For a free indie platform that's a normal stage-of-life gap rather than a red flag, but it does mean the security score rests on hygiene, not on audited assurance. The irony of a prompt-injection trainer with no public vulnerability disclosure channel is worth noting, not condemning.

Accessibility & Growth

Accessibility lands at 92/100 on the Lighthouse audit, which sounds better than it is. The two failing checks are the ones that actually affect users: prohibited ARIA attributes and insufficient color contrast. On a dark theme with light-gray logo rows and muted secondary text, contrast failures are easy to ship and easy to miss. There's no accessibility statement, no WCAG conformance claim, and no mention of keyboard navigation or screen reader support in the collected content, so the automated score is doing all the work here. Growth is the more interesting question. The category is real and growing: prompt injection sits at LLM01 in the OWASP Top 10 for LLM Applications, and every company shipping an LLM feature now has a reason to care. Against the listed alternatives, PromptTrace has a genuine wedge. Gandalf is a single-purpose password-extraction game. HackAPrompt is a competition platform with less interactive feedback. PortSwigger is excellent but web-security-first, not LLM-first. Context Trace, the real-time view of system prompt, RAG context, tool definitions, and user input, is the thing none of them show you, and it's the thing that turns 'I got lucky' into 'I understand why the defense failed'. The limits are honest ones: the site claims 1,223 learners and 7,479 challenges solved, which is early, and the FAQ states plainly there are no paid tiers or enterprise licensing. Free is a great acquisition strategy and a terrible revenue strategy, and the leaderboard-plus-progress-tracking angle for DevSecOps managers is the only visible path to money. The idea has room; the business model is a shrug.

Conclusion

If you're a developer shipping LLM features and you've never actually tried to break one, PromptTrace is the cheapest hour you'll spend this month. The Context Trace alone is worth the sign-in: seeing the system prompt, retrieved documents, and tool definitions laid out next to your input makes prompt injection stop being abstract. Security engineers and red teamers will find the Gauntlet useful for keeping sharp, and the OWASP mapping gives the content a spine that most 'learn AI hacking' content lacks. What it isn't yet is a platform. There's no team management, no certification, no enterprise story, and the FAQ is upfront that there never will be paid tiers. That's fine for a free training ground and a problem for anyone hoping to build a program on top of it. Go in expecting a well-built lab environment and a curriculum that respects your time, and you'll get your money's worth, which is easy since there isn't any. Go in expecting a career platform and you'll be waiting a while.

Suggestions for improvement

Publish a security page with a disclosure policy and bug bounty; a prompt-injection trainer with no vuln channel is an odd look.
Fix the two Lighthouse a11y failures: prohibited ARIA attributes and the low-contrast gray-on-dark logo and secondary text.
Add an accessibility statement and confirm keyboard navigation and screen reader support in the labs, not just the marketing site.
Clarify the path beyond free: team progress tracking, private labs, or certification are the obvious ways to fund the platform.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	PromptTrace	Lakera Gandalf	HackAPrompt	PortSwigger Web Security Academy
Real-time visibility into assembled prompt layers (system prompt, RAG context, tools)	Yes — Context Trace is the core feature, with secrets redacted	No — single-purpose password extraction game	No — competition platform, no live context inspection	No — web security labs, not LLM prompt context
Breadth of hands-on lab content	10 labs, 17 CTF levels, 9 modules, 70+ resources	Single challenge format with escalating levels	Competition challenges plus educational resources	Very broad web security lab catalogue, not LLM-specific
Free access without payment or licensing	Yes — 100% free, no paid tiers or enterprise licensing	Yes — free to play	Yes — free competition platform	Yes — free labs, paid certification exams
OWASP-aligned curriculum mapping	Yes — aligned to OWASP Top 10 for LLM and Agentic Applications	No explicit OWASP mapping	Research-backed, not framed as OWASP curriculum	Maps to OWASP Top 10 for web, not LLM applications
Progress tracking and leaderboards for teams	Yes — progress tracking and leaderboards, no team management	No — individual play only	Competition leaderboards, no ongoing progress tracking	Per-user lab progress tracking, no team dashboards

Lakera Gandalf

A free game where you try to extract a password from an LLM through prompt injection. Fun and accessible, but single-purpose with limited lab variety and no context inspection.

HackAPrompt

A research-backed prompt hacking competition platform with educational resources. Good breadth of challenges, but less interactive feedback and no live Context Trace.

PortSwigger Web Security Academy

Free, high-quality web security labs covering a broad range of vulnerabilities. Excellent depth but not focused on LLM-specific attacks like prompt injection or RAG poisoning.

Comparing options? See PromptTrace alternatives, scored side by side

View all Security