Much cheaper, though still hokey, ideas that you should have already thought of at some point:

• A "formalization office" that checks and formalizes results by alignment researchers. It should not take months for a John Wentworth result to get formalized by someone else.
• Mathopedia.
• Alignment-specific outreach at campuses/conventions with top cybersecurity people.