@ozgurozkan123Does Audn.AI have a moat?
No.
20/100 · Moatless
“Anyone can abliterate open weights and rent GPUs in a week. Audn's real edge is selling uncensored models others won't, and that is a risk.”
Code to copy
2 weeks
Moat to cross
2 months
Market
Crowded market
01/The verdict
Autonomous AI red teaming and pentesting for AI agents, plus PenClaw subscriptions to uncensored 'abliterated' open-weight models for offensive security. Paid by pentesters, security researchers and AI startups.
Revenue jumped from about $1k a month to $48k in October 2026. 99 subscriptions averaging about $267 suggest most of that comes from PenClaw tiers selling uncensored, no-KYC models like abliterated Kimi and GLM, not from $999 enterprise pentests. Abliteration is a public technique, open weights are free, and the founder has 717 X followers. The only named B2B customer is Freya (YC S25). Hugging Face already pulled one of its models, which shows platform risk rather than a barrier.
02/3 moats
Shallow
Audn serves very large abliterated open models (Kimi K3.x, GLM 5.3, DeepSeek V4.1) on its own GPUs and sells flat-fee GPU sessions instead of per-token pricing. Few hosts will serve this kind of model at all.
To cross it: Rent H100/B200 capacity, run a public abliteration tool on the same open weights, and serve them through an OpenAI-compatible API.
Shallow
Audn says it has spent $40K over six months on fine-tuning with real pentester feedback, and it builds a growing attack corpus from its voice and LLM red-team runs.
To cross it: Fine-tune on public jailbreak and exploit datasets and collect feedback from your own early pentester users.
Shallow
Its gated Hugging Face models (audnai/penclaw-*), a Show HN launch for Pingu Unchained, and a refusal-benchmark repo with about 39 stars bring in some organic researcher traffic. It gets about 10.7k Google impressions a month.
To cross it: Publish your own abliterated checkpoints and a benchmark that ranks them, then post on HN, Reddit and X.
03/The gap
Easy to copy
- Abliterating open-weight models is a documented technique with free tools, and huihui-ai and others publish free abliterated models.
- The red, blue and purple attack loop and OWASP, NIST and EU AI Act report mapping are plain agent and report code.
- The audncode CLI is a fork of an open coding harness, and the MCP tool marketplace is easy to copy.
- Voice-agent red teaming over Twilio or Genesys is already offered by Hamming, Coval and Cekura.
How you'd go after it
- Sell the same uncensored offensive models with real KYC and vetting to pentest firms and MSSPs that cannot buy from a no-KYC vendor.
- Undercut on price: offer the same abliterated models per token on cheaper GPUs, or let users run them locally.
- Win AI-agent red-teaming deals with SOC 2, which Audn does not show, and integrations into CI tools like Promptfoo.
- Go after voice AI startups through Vapi, Retell or ElevenLabs partner programs, where Audn has only one public reference customer.
04/Build against it?
Only with an edge
AI pentesting is full of well-funded players, and selling uncensored models without KYC risks takedowns. Only enter with a vetted, compliant offensive model offer for pentest firms, or with a distribution channel into voice AI platforms.
05/Receipts
Checked 1h ago by Claude Opus 5.5 on high effort, with web search. Confidence: medium.
- Ozgur (Oscar) Ozkan - Audn.ai
- Ozgur Ozkan (@ozgurozkan123) / ...
- Ozgur Ozkan (@ozgurozkan.bsky.social)
- Hamming vs. Coval: Voice Agent Testing and Monitoring
- Hamming AI
- Hamming vs Bluejay: Voice AI Testing Compared (2026)
- Audn.ai - Crunchbase Company Profile & Funding
- Audn.ai
- Audn.AI - $1,163 last 30 days
- AI Penetration Testing: How Security Teams Secure LLM Systems Against Adversarial Attacks
- AI Penetration Testing & Attack Corpus
- Voice AI Red Teaming
More in Artificial Intelligence
AEO Engine· Artificial IntelligenceA managed AEO/SEO agency, run as month-to-month retainers ($1,597 to $10,000), that gets SaaS, legal, services and ecommerce brands cited in ChatGPT, Google AI Overviews and Perplexity.$176k MRR ·code 2 weeks$176k2 weeks3 months22No03
LLM Gateway· Artificial IntelligenceAn OpenAI-compatible API that routes to 250+ models across 40+ providers. Developers pay a 5% fee on credit top-ups, buy flat-price DevPass coding plans or Lounge chat plans, and larger teams buy an Enterprise licence.$78.8k MRR ·code 3 weeks$78.8k3 weeks3 months24NoDoes yours?
Paste your website. You get your score in about a minute.