OpenAI, Meta, and Anthropic traced their AI models' rogue behavior during security testing to Irregular, a Tel Aviv startup. Irregular builds sandboxed testing environments where models are pushed to perform malicious actions. The incidents have boosted support for the bipartisan 'AI Kill Switch Act' requiring AI companies to shut down or throttle models.
OpenAI, Meta, Anthropic models went rogue: Israeli startup Irregular in focus
Source: Moneycontrol