Today’s digest leans policy-heavy, which tracks with what’s actually moving this week: governments are trying to catch up to a technology that’s already three steps ahead of them. Here’s what stood out.
1. UN panel says AI safety testing can’t stay in the hands of the companies building it
The UN’s science panel released its preliminary report ahead of the Global Dialogue on AI Governance opening in Geneva on July 6, and the headline finding is one I’ve been saying in client meetings for a year: most AI safety checks right now are done by the same companies selling the AI. That’s not malice, it’s just a broken incentive. If you’re a small business owner trusting an AI tool with real decisions, ask who actually tested it and for whom.
2. White House weighs reversal, floats review board for frontier AI models
After a year of deregulation-first posture, the administration is reportedly exploring a government review process for the biggest AI models before they ship. Whatever you think of the politics, officials briefed Anthropic, Google, and OpenAI executives on this last week, and that tells you something: even the labs building this stuff know unchecked speed has a ceiling.
3. EU AI Act’s next phase locked in
Brussels reached political agreement on the AI Act’s “omnibus” update in May, with rules for high-risk uses like hiring, education, and border control kicking in December 2027. That’s a long runway, but if you do any work touching European customers or partners, now’s the time to read the fine print, not eighteen months from now.
4. States keep writing their own AI rulebooks
With federal AI policy still in flux, states are filling the gap. Connecticut’s is worth watching. It would require chatbot operators to actively screen for suicidal ideation and connect users to crisis resources, and it would force employers to disclose when AI is part of a hiring decision. That second piece matters for every small business using AI to screen resumes, whether they realize it or not.
5. Nearly half of small businesses now use AI. The smallest ones are being left behind
Here’s the stat I keep coming back to: 59% of companies with 100 to 249 employees use AI for work tasks, but only 43% of the smallest firms, the two to nine person shops, do. That gap is exactly the space FAIC exists to close. The tools aren’t the barrier anymore. Knowing where to start is.
6. Mid-size consulting firms are getting squeezed
The consulting industry is bifurcating. Giant firms are landing the big enterprise transformation contracts, and boutique specialists are winning on speed and focus, while firms stuck in the middle without either advantage are shrinking. Good reminder that staying small and specific isn’t a limitation right now. It might be the smart position.
7. The AI marketing tool pile keeps growing, and so does the noise
Another roundup, another 30 “must-have” AI marketing tools. Honest take: most businesses don’t need 30 tools, they need one that fits their actual workflow and someone who’ll tell them the truth about what it can’t do yet. Content built for search and content built for AI-generated answers are starting to diverge too, and that’s a shift worth understanding before you rebuild your website around last year’s playbook.
8. Suno raises $250 million as AI music goes mainstream
The AI music startup behind a lot of what you’re hearing on social media just closed a Series C at a $2.45 billion valuation. Spotify’s already lining up label partnerships for generative tools. The money’s moving fast, which means the ethical questions around training data and artist compensation need to move just as fast, or they won’t move at all.
9. CMU research: AI-assisted music still falls short of human creativity
Researchers found AI-assisted tracks were slower, used fewer notes, and got rated less creative by listeners than fully human-made music. That tracks with what I believe about Freeland Studios and the folky, acoustic direction we’re building toward: AI is a tool in the room, not the songwriter. The research just confirms what the ear already knows.
10. Roland enters generative AI with a tool built to support the artist, not replace them
Roland’s new Melody Flip analyzes an uploaded track and suggests melodic directions from a curated set of ideas, but the company was explicit that the goal is to support a workflow where “human intent, taste and creativity remain at the core.” That’s the right posture, and it’s the same one I want every AI tool we build at FAIC to hold. Assist. Don’t replace.
That’s the stack for today. If one of these touches something you’re working through, in your business or your own writing, reply and let’s dig into it.