Attune 1.0
AI model by Abnormal AI
The Foundational Behavioral AI Model for Security
It powers 100% of detections on the Abnormal platform by baselining normal behavior to catch deviations like BEC and account takeover.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
Category: Machine Learning
- License
- Proprietary
- Access
- Integrated into Abnormal Behavioral Security Platform
- Training Data Size
- 10 Billion+ behavioral signals
- Detection Power
- 100% of malicious detections
- Unique Attack Discovery Improvement
- 50% more per week
- Data History
- 8 years of proprietary data
- Powers 100% of malicious attack detections
- 50% more unique attacks discovered per week
- Built on 8 years of proprietary behavioral data
Overview
Attune 1.0 is Abnormal AI's foundational behavioral AI model designed to power the Abnormal Behavioral Security Platform. By building these baselines, the model can detect meaningful deviations that signal sophisticated attacks like business email compromise and account takeover. Attune 1.0 operates at machine speed to investigate and respond to threats without relying on traditional signatures or indicators of compromise. It is built specifically for security applications, delivering higher detection efficacy and fewer false positives than general-purpose LLMs. The model leverages eight years of proprietary behavioral data to protect email, identity, and AI environments.
Behavioral Intelligence
- Known-Good Baselining: It builds a baseline of known-good behavior for every employee and vendor based on communication patterns and sign-in events.
- Natural Language Processing: The model applies natural language processing to detect abnormalities in email behavior that indicate potential attacks.
- Multi-Signal Ingestion: It ingests thousands of diverse signals via API to understand work, access, and relationships across email, identity, and cloud platforms.
Detection Efficacy
- 100% Detection Power: Attune 1.0 powers 100% of all malicious attack detections within the Abnormal Behavioral Security Platform.
- Unique Attack Discovery Improvement: The model discovers 50% more unique attacks per week compared to prior model generations.
- Signature-less Detection: It is engineered specifically for behavioral detection rather than relying on traditional threat signatures or indicators of compromise.
- Real-World Training: The model is sharpened on real-world attacks to improve its ability to catch novel threats that general-purpose LLMs might miss.
Use cases
- Security teams use Attune 1.0 to detect business email compromise by identifying deviations in communication patterns.
- Organizations protect cloud applications by using the model to catch account takeover attempts through behavioral anomalies.
- SOC analysts reduce manual workload by relying on the model to autonomously investigate and remediate threats at machine speed.
- IT leaders secure non-human identities like AI agents and service accounts by mapping their behavior in the Identity Trust Graph.
Specifications
- Modalities
- Behavioral signals, natural language → threat detection and evidence
- License
- Proprietary
- Access
- Integrated into Abnormal Behavioral Security Platform
- Training Data Size
- 10 Billion+ behavioral signals
- Detection Power
- 100% of malicious detections
- Unique Attack Discovery Improvement
- 50% more per week
- Data History
- 8 years of proprietary data
- Identity Coverage
- Human, Non-Human, and Cloud Identities
- Integration Method
- API-based ingestion
How it works
- Data Ingestion and Mapping: The model ingests thousands of behavioral signals via API from email, identity providers, and cloud platforms to map every human and non-human identity.
- Behavioral Baselining: Attune 1.0 builds a baseline of known-good behavior for every identity, learning their unique communication patterns, access habits, and relationships.
- Anomaly Detection: The model recognizes meaningful deviations in context, identifying sophisticated attacks like business email compromise and account takeover without signatures.
- Autonomous Response: Operating at machine speed, the model investigates threats, takes remediation actions, and provides evidence for security teams.
FAQ
What makes Attune 1.0 different from general-purpose AI models?
Attune 1.0 is a foundational AI model built specifically for security, trained on eight years of proprietary behavioral data. Unlike general-purpose LLMs, it is architected to understand the behavior behind the risk for people, AI agents, and cloud identities.
How effective is Attune 1.0 at detecting attacks?
Attune 1.0 powers 100% of all malicious attack detections within the Abnormal platform. It has demonstrated a 50% improvement in discovering unique attacks per week compared to previous model generations.
What platforms does Attune 1.0 integrate with?
The model ingests thousands of behavioral signals via API from sources including Microsoft 365, Google Workspace, Slack, Entra, Ping Identity, GCP, and Azure. It maps these into an Identity Trust Graph covering human and non-human identities.
Added by AntSpace from public sources (abnormal.ai).