Security tools were built for human attackers and human analysts. Both assumptions are breaking: attackers now run swarms of AI agents that retry and rewrite themselves faster than anyone can watch, and companies hand their own AI agents real credentials to code, operate and act on their behalf.
We think defense has to move to where those agents run. Solomon decides on the system call itself, inside the Linux kernel, so containment happens before lateral movement. Above it, AI agents analyze every signal and run the investigation, and every action they propose first passes deterministic rules and, where it matters, a human who approves it with a hardware-signed tap.
Kernel acts. AI agents think. Humans decide. That loop is the company.