“A major bottleneck for improving defensive AI cybersecurity models is the lack of public data, as companies typically do not share logs from security incidents.”