Alice Raises $140M to Stress‑Test AI Models from an Attacker’s Point of View

U.S. startup Alice, which specializes in testing AI systems for safety and security, has raised a $140 million round led by Apax Digital Funds, the company disclosed on August 25. Alice runs large batteries of adversarial prompts and autonomous agent tasks against leading large language models to uncover jailbreaks, prompt‑injection vectors and other failure modes before the models are widely deployed. Its customers include major labs such as Anthropic, Google and Cohere.

The founders draw on nearly a decade of experience policing abuse and spam at internet‑scale platforms, experience they say gives them a realistic view of how determined attackers will try to bend AI to their purposes. Alice’s platform aims to industrialize “red‑teaming” — the kind of simulated offensive testing that regulators increasingly expect model providers to perform and document as part of their risk management.

As governments move toward tighter oversight of powerful AI systems, labs and cloud providers need credible, third‑party evidence that they have probed for dangerous capabilities. With fresh capital earmarked for R&D and hiring, Alice is positioning itself as a core part of the AI safety stack: a company whose job is to break models before the real world does.