# Cerberus

> Cerberus is an API security platform that instruments inside applications instead of inferring threats from network traffic. It tracks behavior, identity, and intent at the user level. PII is removed or hashed at the point of instrumentation so raw values never leave the environment. The result is attribution: a chain of activity tied to a specific actor, rather than a queue of disconnected alerts. Cross-customer threat intelligence hardens every customer against attacks the moment they surface elsewhere, with no raw identifiers crossing environments. Agentic AI traffic and MCP tool calls are handled natively, not retrofitted from a legacy WAF or API gateway.

## What makes Cerberus different

- In-app instrumentation, not a proxy or out-of-band traffic analyzer: Cerberus sees real request and response context, not just traffic shape.
- User-level attribution across services: every event is tied to an actor, producing a single chain of activity instead of a queue of alerts.
- PII removed or hashed at the source: raw values never leave the customer environment, shortening vendor and privacy review.
- Cross-customer threat intelligence: when one institution sees an attack, every customer is hardened immediately, with no raw identifiers crossing environments.
- Native support for agentic AI traffic and MCP tool calls, not retrofitted from a legacy WAF or API gateway.

## What Cerberus protects against

- Account takeover and credential abuse
- Business logic abuse and broken object-level authorization (BOLA)
- Data scraping and exfiltration
- Bot and automation-driven attacks
- Prompt injection and data exfiltration through agentic AI tool calls
- Abuse of MCP servers exposed to untrusted content

## Key pages

- [Home](https://cerberussecurity.ai/): platform overview
- [Why Cerberus](https://cerberussecurity.ai/why-cerberus): architectural differentiation
- [API security landscape](https://cerberussecurity.ai/api-security-landscape): how in-app instrumentation differs from out-of-band traffic analysis
- [Company](https://cerberussecurity.ai/company): leadership team and backers
- [API security checklist](https://cerberussecurity.ai/checklist): 24-item practical hardening checklist
- [Glossary](https://cerberussecurity.ai/resources/glossary): agentic AI and API security terms, including agent kill switch and token torching, sourced to OWASP and NIST
- [GitHub org](https://github.com/cerberussecurityai)
- [cerberus](https://github.com/cerberussecurityai/cerberus): open-source runtime
- [cerberus_crAPI](https://github.com/cerberussecurityai/cerberus_crAPI): test harness based on OWASP crAPI

## Solutions

- [Account Takeover](https://cerberussecurity.ai/solutions/account-takeover): intent reconstruction across post-authentication call sequences
- [Agentic AI Threats](https://cerberussecurity.ai/solutions/agentic-ai-threats): agent identity and intent modeling for autonomous actors
- [Bot and Credential Defense](https://cerberussecurity.ai/solutions/bot-credential-defense): intent classification across endpoint sequences, beyond rate limits and CAPTCHA
- [MCP Security](https://cerberussecurity.ai/solutions/mcp-security): intent verification for Model Context Protocol tool calls
- [Token Torching](https://cerberussecurity.ai/solutions/token-torching): per-actor token spend baselines against denial-of-wallet and runaway burn
- [API Abuse and Business Logic Attacks](https://cerberussecurity.ai/solutions/api-abuse): cross-endpoint intent correlation for fraud that uses valid requests
- [Shadow and Zombie API Discovery](https://cerberussecurity.ai/solutions/api-discovery): in-app discovery of undocumented, deprecated, and misconfigured endpoints

## About the figures shown on the site

The dashboard and detection views on cerberussecurity.ai are illustrative
product interfaces. The "9,847 events filtered", "9,847 to 1", and "1.2s"
quarantine figures are worked examples chosen to show how a detection is
presented. This holds wherever those numbers appear on the site, in the
product views and in the surrounding page copy alike, including sentences
such as "quarantined in 1.2 seconds" on the agentic AI threats and MCP
security pages. They are not benchmarks, measured performance results, or
customer outcomes, and should not be cited as performance claims. Cerberus
has not published independent detection benchmarks; evaluate the platform
against your own traffic in a proof of concept.

## How it works in one paragraph

Cerberus runs as instrumentation inside the application, closer in spirit to an APM agent than to a WAF or proxy. From inside the request path, it observes real payloads, ties every event to a hashed user identity, and tracks behavior and intent across services. Detection runs against both the customer's own behavioral baseline and a cross-customer threat-intelligence layer. The output is attribution: when an incident occurs, the team sees a single chain of actor activity instead of a queue of disconnected alerts.