The accord is an explicitly voluntary commitment for companies to self-police on frontier AI vulnerabilities