Early access to frontier AI capabilities can give trusted defenders time to find and mitigate threats before those capabilities spread.

A developer that discovers a dangerous capability before it is widely available can put it to defensive use with appropriate partners. In cyber security, an agent that helps find exploitable weaknesses can also help defenders fix them before a similar capability reaches attackers. The advantage depends on the developer sharing useful defensive access and findings while controlling access that would make misuse easier.

Anthropic's Project Glasswing gave selected infrastructure and open-source maintainers access to a frontier model for defensive vulnerability discovery. A UK government pilot using frontier models across public services reported hundreds of findings and remediation of all findings it classified as critical. These examples establish that capable AI is already being applied to defensive work; they do not prove that a particular future attack was prevented or quantify the benefit of being first.

The same reasoning could apply to robotic threats if responsible developers can identify dangerous capabilities and help counter them before harmful use spreads. The evidence for that application is currently thinner than the evidence for cyber defence. The case for a timing advantage therefore rests most firmly on cyber security today.