OpenAI restricted access to cybersecurity-dedicated tools within its model line in the period.
The decision anticipates, on the company's own initiative, the criterion that would guide regulatory decisions in the following weeks. The capability to find flaws in software is the same one used to exploit them, and that symmetry motivates both self-regulation and pre-release review.
Restricting yourself has a concrete commercial advantage over being restricted by someone else: it preserves predictability. A company already limiting access can argue the control exists and works, which changes the tone of the conversation with regulators.
The other side of the argument would surface forcefully the following month, when more than a hundred security leaders signed a petition against models being pulled by government decision, holding that defenders need the same capability.
For anyone working in defensive security, the practical consequence is the familiar one: the restriction reaches those using an identified service, and doesn't reach anyone running open-weights models on their own hardware.