SentinelOne: OpenAI Daybreak cuts malware analysis to eight hours

SentinelOne's starting point was a capacity problem, not a technology one. Stewart says his research teams could reverse only very small numbers of pieces of software, because deep investigation had to be reserved for the most interesting or important threats. With fewer samples to compare, there were fewer chances to notice connections between them.

OpenAI Daybreak changes that arithmetic by giving defenders access to OpenAI models for cybersecurity work. If reverse engineering that once took four to six weeks takes about eight hours at human-level output, researchers can point that time at additional samples. A component appearing in two apparently unrelated samples can then become a lead during a customer investigation, and comparing findings across more samples can surface broader relationships between threats.

Stewart frames the payoff in terms of reach rather than raw speed: mapping a much larger part of the malware landscape and defending further out. He draws one boundary — accountability stays with his team, and architecture-changing decisions still warrant human review, while point fixes may be automated. He also wants the same capabilities inside code creation, so developers can handle security issues as they build.