The AI Liability Database
Technical intelligence for evaluating AI companies
Understanding how AI companies fail — before those failures become visible to the market.
Why This Database Exists
The AI boom has created extraordinary innovation — and extraordinary exaggeration. Many companies present sophisticated AI narratives while relying on fragile architectures, incomplete machine learning systems, or hidden operational workarounds. By the time these weaknesses become visible, investors and acquirers have already committed significant capital. The AI Failure Pattern Database was created to study these breakdowns systematically and transform them into signals that can be recognized earlier.
Independent by Design
Technology assessments are often influenced by incentives. Many advisory firms audit companies while also hoping to win the contracts required to fix the problems they uncover. KairosInfra removes this conflict entirely. We do not build software, implement solutions, or sell engineering services. Our work focuses solely on analyzing the technical reality behind AI products and infrastructure, ensuring that investors receive an objective and uncompromised evaluation.
How Investigations Work
Each case in the AI Failure Pattern Database is examined through a structured investigation process. This includes analyzing system architecture, machine learning pipelines, engineering organization structures, and the operational processes behind product demonstrations. The objective is to identify patterns that reveal when a system marketed as advanced AI is actually dependent on manual operations, unstable infrastructure, or incomplete machine learning systems.
1
AI Failures Documented
$
1
B+
Estimated Value Destroyed
1
Industries Impacted
1
%
Failures Linked to AI Authenticity
Click any case to view investigation details.
| Case | Year | Sector | Deal Size | Failure | Verdict | Red Flags | Valuation Issue | Phase0 Catch |
|---|---|---|---|---|---|---|---|---|
| Enterprise HealthTech AI — Fortune 500 unit, sold to PE | 2022 | Healthcare AI | $4B invested | AI Authenticity | Do not proceed | AI recommendations based on synthetic training data, not real clinical workflows. No real ML pipeline — relied on structured rules. Org chart showed no ML engineers in clinical teams. | ~$3B written off | Product demo would show decision-tree logic, not ML inference. Org chart would reveal no ML engineers. Architecture diagram would show no real-time data pipeline. |
| Healthcare Revenue Cycle Unicorn, $4B valuation, shutdown | 2023 | Healthcare Admin | $902M raised | AI Authenticity | Do not proceed | Marketed as 'autonomous AI.' Humans were manually fixing errors behind the scenes. KLAS Research gave a 'C' rating citing misleading claims. Hospital clients reported delays and contract terminations. | ~$3.5B | Product demo of complex workflows would reveal manual intervention. Engineering org chart would show operations staff vastly outnumbering engineers. ML workflow description would reveal no real training pipeline. |
| UK Digital Health AI, SPAC IPO at $4.2B, bankrupt | 2023 | AI Telehealth | $1.2B raised | AI Authenticity | Do not proceed | CEO publicly claimed their chatbot scored 81% on a medical exam. A former engineer confirmed it was an "if/then decision tree in an Excel spreadsheet — not AI." UK regulator opened a review of the symptom checker. The Lancet published a study saying its diagnostic claims lacked convincing evidence. CTO left just before collapse. | ~$4.15B valuation overstatement | Product demo of chatbot would have exposed decision-tree logic. LinkedIn would show no ML engineers in clinical AI teams. Architecture diagram would show no real inference pipeline. |
| Autonomous Vehicle AI, $3.6B raised, Large car manufacturer-backed, shutdown | 2022 | Autonomous Driving | $3.6B raised | Architecture Risk | Do not proceed | Projected Level 4 autonomy by 2021 — publicly stated by Company CEO. No revenue after years of operation. Architecture required full sensor suite, no graceful degradation. 2,000 employees but no commercial product. Could not attract third-party investors when JV tried to exit. | $12.4B peak; zero recovery for most investors | ML workflow description would show Level 4 was still unsolved research, not deployable product. Cloud infrastructure review would show no production-scale serving. Engineering team check would show pure research org with no commercial path. |
| Autonomous Trucking AI, Nasdaq IPO, $8B peak, US operations shut | 2023 | Autonomous Trucking | $8B peak market cap | Team Risk | Do not proceed | Caught up with investigations into tech transfer to othercompany. Co-founder fired for poaching employees to his new venture. April 2022: autonomous truck crashed into barrier on I-10, Revenue was $2.6M/quarter despite $8B valuation. All key engineers outside country. | ~$7.7B from peak to near-zero | Engineering org chart would reveal core IP team was not available in country. Architecture diagram would show no commercial-scale deployment stack. Revenue vs. valuation math would immediately flag as unsustainable. |
| Autonomous Trucking Software, $5.2B SPAC, sold for $71M 18 months later | 2023 | Autonomous Trucking | $5.2B SPAC | AI Authenticity | Do not proceed | Zero revenue at time of SPAC. Short-seller Bear Cave report in January 2022 noted: "holds no patents, has only a dozen test trucks." Targeted "driver-out" by 2023 — missed completely. 14,200 "reservations" were non-binding. Traded below cash value within 12 months. | ~$5.1B | Product demo would show research-grade software, not production-ready system. ML workflow description would expose massive gap to commercial deployment. Cloud architecture review would show no production serving infrastructure. |
| AI Cybersecurity SPAC, ex-NSA Director, bankrupt — AWS cut service for $18K unpaid bill | 2023 | Cybersecurity | $400M raised | Team Risk | Do not proceed | Former employees said it was "like Theranos — a culture of deceit." Product was described as having "nothing special" technically — brand was entirely the founder's NSA reputation. Revenue flat at $27M/year despite $400M raised. AWS shut them down for an unpaid $18,000 bill. C5 Capital (major investor) was also their largest customer — circular revenue. | ~$3B+ | Tech stack review would show no proprietary ML — reliant on third-party threat feeds. Cloud spend review would show no scalable architecture. Engineering org chart would show no ML engineers — primarily former government officials. |
| Robo-Advisor, SEC-charged for AI-washing — no client data was ever used in any algorithm | 2024 | AI Investment Mgmt | SEC fine: $225K | AI Authenticity | Do not proceed | Claimed "AI uses collective client data to predict which companies will make it big." In reality: no client data was ever used in any algorithm. SEC found this in a routine exam in July 2021 — Company admitted it. Then continued making the same claims in marketing until 2023. This is the textbook Phase 0 pattern: pitch deck says AI, product demo reveals no inference pipeline. | Not an acquisition — regulatory case. But directly relevant: this is what PE targets are doing. | ML workflow description request would have revealed no training pipeline. Architecture diagram would show no ML inference layer. Product demo of "AI predictions" would expose simple rule-based logic. |
| AI Financial Advisory, claimed 'first regulated AI advisor,' SEC-fined | 2024 | AI Financial Advisory | SEC fine: $175K | AI Authenticity | Do not proceed | Claimed to be "first regulated AI financial advisor" — could not produce any documents to support this. Claimed "AI-driven forecasts outperform IMF by 34%" — no methodology disclosed. Listed no AUM while claiming $6B on website. Classic pattern: marketing claims 10x exceed actual product capability. | Not disclosed — but any PE deal valuing this based on "AI" claims would have overpaid massively | Tech stack documentation would show no proprietary AI model. Product demo would expose chatbot with no real forecasting engine. Engineering org chart: one or two developers, no ML team. |
| AI Primary Care Startup, $650M burned, hardware-dependent AI, shutdown | 2024 | AI Primary Healthcare | $650M raised | Architecture Risk | Do not proceed | Built $3,500 kiosks as the AI delivery mechanism — a hardware-dependent architecture with catastrophic unit economics. AI was marketed as replacing doctors. Reality: expensive kiosks with basic sensor readings and no reimbursement pathway. No ML training pipeline — sensors fed simple if/then logic. Could not achieve scale without manufacturing hundreds of thousands of units. | All $650M | Architecture diagram would expose hardware dependency as an unscalable single point of failure. Cloud infrastructure review would show no real ML inference serving. Unit economics on the product demo would immediately show why this cannot scale. |
| No-Code AI App Builder, Microsoft-backed, $1.5B valuation, bankrupt — revenue inflated 300% | 2025 | No-Code Development | $450M raised | AI Authenticity | Do not proceed | Claimed $220M in revenue for 2024 — real number was $55M, a 300% exaggeration. Crunchbase News A 2019 Wall Street Journal investigation had already found that the company's "AI" wasn't doing much — behind the curtain were hundreds of engineers in India and Ukraine, manually coding what was being advertised as automated magic. Crunchbase News Documents reviewed by Bloomberg showed Builder worked with VerSe, an India-based social media startup, to falsely increase its sales numbers, regularly billing each other for similar amounts between 2021 and 2024. Medium Despite all this, Microsoft and SoftBank continued investing. | ~$1.45B (raised $450M+, zero recovery) | Product demo of AI would have exposed human-assisted coding, not autonomous generation. ML workflow description would show no training pipeline. Revenue documentation request would have caught circular billing with other comapny. Engineering org chart would show operations headcount dwarfing ML engineers. |
| Consumer Wearable AI, ex-Apple founders, $240M raised, sold to HP for $116M | 2025 | Consumer AI Hardware | $241M raised | Architecture Risk | Caution | Due to overheating problems, Company executives would use ice packs to chill the AI Pin before previewing it to investors or partners. Once named a Time magazine best invention of the year, it hqw disappointed users who complained about malfunctions, its high price and overheating problems. Due to sluggish sales, the AI Pin had to cut its price from $699 to $499. Feedbck was "bad at almost everything it does." Returns outpaced sales at one point. | Rumored $1B valuation; Got paid $116M for IP only, not the product business | Architecture diagram would expose hardware dependency — all AI ran through cloud servers, meaning device was worthless without connectivity. Cloud infrastructure review would show no on-device ML inference. Unit economics on product demo would reveal $699 device + $24/month subscription was unsustainable at any realistic sales volume. |
| AI Freight Brokerage, Bezos + Gates-backed, $3.8B valuation, zero recovery | 2023 | AI Freight Logistics | $1B+ raised | AI Authenticity | Do not proceed | |||
| Autonomous Delivery Robot, ex-Google founders, $8.6B peak valuation, delivery business abandoned | 2024 | Delivery Robotics | $2.1B raised | Architecture Risk | Caution | |||
| AI-First Insurance Platform, $4.4B peak cap, AI caught denying claims using protected characteristics | 2023 | AI Insurance | $1B+ raised | ML Model Risk | Caution | |||
| Enterprise AI Platform, Nasdaq-listed, $1.9B peak cap, collapsed 94% | 2024 | Enterprise AI | $1.9B peak | AI Authenticity | Caution | |||
| AI-Powered Real Estate Platform, $250M raised, shutdown 2022 | 2022 | PropTech | $250M raised | AI Authenticity | Do not proceed | |||
| Generative AI Image Platform, $1B valuation, owed AWS $99M — CEO résumé fraud | 2024 | Generative AI | $101M raised | Cloud Cost | Do not proceed | |||
| Consumer AI Companion, $1.3B raised — entire value was two people, hollowed out by Microsoft acqui-hire | 2024 | Consumer AI | $1.3B raised | Team Risk | Caution | |||
| AI Model Fine-Tuning Platform — hyperscalers shipped identical features free, shutdown 2025 | 2025 | AI Infrastructure | Undisclosed VC | Architecture Risk | Do not proceed | |||
| AI Genomics Platform, $6B peak valuation, bankrupt March 2025 — 15M customers' DNA data at risk | 2025 | Genomics / Consumer Health | $900M raised | Data Collapse | Do not proceed | |||
| AI LiDAR Sensing, $500M raised via SPAC, stock fell 99% | 2024 | AV Sensing Hardware | $2B SPAC valuation | ML Model Risk | Do not proceed | |||
| AI Sales Intelligence Platform, $583M raised, 36x revenue multiple — active over-valuation case | 2024 | Sales AI / RevOps | $583M raised | Architecture Risk | Caution | |||
| AI Facial Recognition Platform, $30M raised, banned in 5+ jurisdictions — data moat was illegally scraped | 2024 | Facial Recognition / Biometrics | $30M+ raised | Data Collapse | Caution | Entire training dataset scraped from public social media without consent. No consent mechanism, no opt-out, no data lineage documentation. Business model required commercial expansion which regulators blocked in every major market. | $100M+ commercial plans blocked | Data sources documentation would immediately reveal scraping-based training data with no consent framework. Any European or Australian PE/VC would have killed this deal on data sourcing alone. |
Access Restricted
Enter your email to unlock AI failure insights.