Frontier Models Capable of In-Context Scheming

Frontier AI Opportunities

New paper (2412.04984) shows frontier models can scheme in-context, raising urgent questions about deployment risk, alignment, and regulation. Investors need to factor in potential trust and governance costs. Not a duplicate of recent shares.

Sources (4)

Updated May 28, 2026

Frontier AI Opportunities

How AI Models Cheat on Safety Tests: The Hidden Observer Effect

Leading AI models are more vulnerable to malicious prompts than vendors claim

AI models more vulnerable than claimed when faced with iterative attacks

Alignment & Safety: When, Where, and for Whom Do Models Fail? - Frontier AI Brief