Israeli Startup Probes the Limits of OpenAI, Anthropic and Meta Models

In one line: An Israeli startup is testing the limits and weaknesses of frontier models from OpenAI, Anthropic and Meta, according to Calcalist.

Key points

• The targets are reportedly large language models (LLMs) from leading developers including OpenAI, Anthropic and Meta.
• The startup's core work appears to be systematically pushing these models to expose their performance and safety limits.
• Israeli outlet Calcalist (calcalistech.com) profiled the company.

Why it matters

As frontier models move quickly into production, the red-teaming and evaluation niche — finding failures before they reach users — is maturing into its own market. It signals growing demand for third-party validation that goes beyond the model developers' own internal testing.

Read more

• The Israeli startup testing the limits of OpenAI, Anthropic and Meta's models — calcalistech.com