OpenAI Readies Rollout of New Cyber Model as Industry Shifts to Defense

The article reports that major AI providers are treating advanced models as potentially dangerous, implementing restricted rollouts and vetted-access programs (e.g., OpenAI’s “Trusted Access for Cyber” pilot and Anthropic’s Mythos Preview) to accelerate defensive research while limiting weaponization. Industry leaders caution that such capabilities will likely leak or be reproduced, and the move reflects a shift from consumer releases toward guarded distribution and responsible disclosure-like practices.