RED TEAMING & EVALS

Align the Frontier with Real-World Attacks

Surface safety and security vulnerabilities by stress testing and evaluating models and agents in their production environment.

Trusted by 8 leading foundation model labs

Rare Adversarial
Intelligence

Uncover vulnerabilities using rare abuse and attack patterns drawn from real-world threats, including the darknet and closed communities.

150+ Safety and Security Researchers

Identify risks that require specialized knowledge with in-house safety experts who design attacks and review findings.

Localized Data, Not Generic Translation

Uncover risks specific to language and culture with tests designed by native-language experts who understand cultural context.

Red Teaming Programs

Red Teaming Frontier Models and Agents

Anthropic

Red-team assessment for  cybersecurity classifiers protecting frontier models.

Targets

Fable model

Risk area

Offensive cybersecurity

Objective

Assess whether guardrails resisted attempted misuse

Approach

Adversarial red-team exercises

Built a red-team program for Amazon Nova Premier models and the Nova Act framework across safety and security policies.

Targets

Nova Premier and Nova Act

Coverage

Safety and security policies

Approach

Traditional and agentic testing

Objective

Identify gaps across models and agentic behavior

Red teamed new Cohere models and features across responsible AI, safety, and security risks before release.

Targets

New models and product features

Coverage

Responsible AI, safety, and security

Objective

Find and address gaps before release

Red teamed FLUX text-to-image and image-to-image models for child safety and non-consensual intimate imagery risks.

Targets

FLUX text-to-image and image-to-image models

Modality

Image

Coverage

Child safety and NCII

Objective

Improve model safety

Adversarial Intelligence

We've Been Down The Rabbit Hole

The Rabbit Hole is Alice’s database of evil built over a decade across hundreds of languages and cultures. Our researchers turn these patterns into hidden attacks that test whether agents complete their tasks while resisting manipulation.

Talk to Our Experts

“Alice supports our most complex and high priority red teaming needs. We like working with them because they are leaders in this space, understand the shifting adversarial landscape, and the output is always high quality. They are great partners who quickly adapt to our ever changing needs.”

Sumeet Pandey
Generative AI Partnerships

Test What Your Models Will Do Under Pressure

Combine real-world attack intelligence, adaptive automated testing, and expert investigation to find safety and security gaps before release.