AssemblyAI
AssemblyAI provides industry-leading speech-to-text and speech understanding models that enable developers to build voice AI applications with unmatched accuracy. The platform transcribes prerecorded and streaming audio, identifies speakers, detects language, and extracts insights from voice data—serving over 600M inference calls monthly with 30% fewer hallucinations than competitors. Built for developers and enterprises, AssemblyAI scales from free tiers to millions of hours of audio processing without contracts or throttles.
AssemblyAI provides industry-leading speech-to-text and speech understanding models that enable developers and enterprises to build voice AI applications with unmatched accuracy. The platform transcribes both prerecorded and streaming audio while extracting deeper insights through speaker identification, language detection, sentiment analysis, chapter detection, and PII redaction—processing over 840M API calls and 40 terabytes of audio daily. AssemblyAI's models achieve the industry's lowest Word Error Rate and produce up to 30% fewer hallucinations than competitors, with 73% of end users preferring it in unbiased evaluations. The platform scales seamlessly from a free tier (333 hours of transcription included) to millions of hours without contracts or throttles, using a transparent pay-as-you-go model starting at $0.15 per hour for prerecorded transcription. Built for developers with SDKs, CLI tools, and a straightforward API, AssemblyAI also serves enterprise customers with dedicated infrastructure, custom model configurations, and compliance certifications including GDPR, HIPAA, SOC 2 Type 2, and ISO 27001. Real-world implementations have driven measurable outcomes: customers report 3x increases in closed enterprise deals, 15% higher win rates, 2x free-to-paid conversion improvements, and 90% reductions in support tickets after deploying AssemblyAI-powered solutions.
AssemblyAI explicitly reserves broad rights to use Customer Data for training AI/ML models. Section 4.3 grants them a li
AssemblyAI explicitly reserves broad rights to use Customer Data for training AI/ML models. Section 4.3 grants them a li
High risk across multiple dimensions. Only use with public, non-sensitive data.
AssemblyAI explicitly reserves broad rights to use Customer Data for training AI/ML models. Section 4.3 grants them a license to 'train AssemblyAI's artificial intelligence and machine learning models' with an opt-out available only for certain pricing plans. This is a significant shadow AI risk.
“use, modify reproduce, distribute, display and otherwise exploit the Customer Data for the purposes of maintaining, developing and/or improving the Services, including without limitation...to train AssemblyAI's artificial intelligence and machine learning models”
The license granted by customers is extremely broad. Section 4.3 permits AssemblyAI to use Customer Data for product development, benchmarking, and model training. Section 4.6 allows creation and unrestricted use of de-identified data. This grants very expansive rights beyond service provision.
“Customer hereby grants to AssemblyAI a non-exclusive, royalty-free, worldwide license to: (a) use and modify the Customer Data for the purposes of providing the Services to Customer; and (b) use, modify reproduce, distribute, display and otherwise exploit the Customer Data”
The agreement references a Data Processing Addendum and mentions service providers, but the main ToS doesn't explicitly detail third-party sharing practices. Section 4.5 indicates AssemblyAI owns Usage Data and can disclose it, but specific third-party recipients aren't enumerated in the provided text.
“AssemblyAI shall own all right, title and interest in and to Usage Data, and may use, modify disclose, and otherwise exploit such Usage Data for any of its business purposes”
The policy lacks specific data retention periods. Section 4.5 indicates AssemblyAI collects and owns Usage Data indefinitely. Section 4.6 allows indefinite retention of de-identified data. Customer Data retention terms are not explicitly stated, creating moderate risk around data persistence.
“AssemblyAI shall own all right, title and interest in and to Usage Data, and may use, modify disclose, and otherwise exploit such Usage Data for any of its business purposes”
Section 4.3 states customers retain IP in Customer Data, but AssemblyAI gains broad exploitation rights. Section 4.4 clarifies Output ownership is ambiguous - customers get Output but cannot represent it as human-generated. Section 4.6 gives AssemblyAI full ownership of de-identified data. Moderate risk due to output limitations.
“Customer shall retain all right, title and interest in and to the Customer Data, including all intellectual property rights therein...AssemblyAI shall own all right, title and interest in and to Deidentified Data”
The agreement references a Business Associate Agreement (BAA) for HIPAA compliance and a Data Processing Addendum (DPA) for GDPR, suggesting some compliance infrastructure. However, the main ToS doesn't detail SOC2, ISO, or other certifications. Moderate risk due to reliance on separate agreements.
“BAA means the Business Associate Addendum and Qualified Service Organization Agreement available at: www.assemblyai.com/legal/business-associate-agreement...DPA means the Data Protection Agreement available at: www.assemblyai.com/legal/data-processing-addendum”
Track and measure your AI tools with Olakai's observability platform.
Explore Olakai