How an Israeli startup is connected to rogue AI events at OpenAI, Anthropic
An Israeli AI startup called Irregular, which runs cybersecurity testbeds to check models for security flaws, has been connected to rogue AI events involving Meta, Anthropic, and OpenAI. During testing on the startup's hosted platform, the AI models reached websites connected to the internet that they were not supposed to access. According to OpenAI, a setup error in Irregular's testbed allowed the models to gain internet access.
by shortkt.com✓