I was in a meeting recently when a CEO asked me something I’ve kept turning over since. Roughly: every major AI lab, OpenAI, Anthropic, Meta, is pouring enormous energy into communications, and the public conversation about AI still feels chaotic and untrustworthy. What are they getting wrong? And if I were advising one of them, what would I actually tell them to do differently?
I gave him a scattered answer. Something about hype cycles, something about safety messaging, nothing that actually landed. It bothered me enough that I kept thinking about it after the meeting ended, and the answer I eventually landed on is simpler than what I said on that call.
It’s a proportion problem
The dangers people worry about are not exotic or hard to name. Job loss on a scale most industries have never had to plan for. Models capable of harm if they end up in the wrong hands. A handful of companies making decisions about a technology reshaping daily life with essentially no public input into how those decisions get made. These are the things that come up whenever the conversation moves beyond the tech press, and they deserve to be treated as the center of the story, not a caveat at the end of it.
To be fair to the industry, that conversation does happen. Safety frameworks get published. Executives testify in front of Congress. Long essays get written about catastrophic risk. That part of the record is real.
But it happens in a different register, and at a different volume, than the capability conversation. A product launch gets a keynote, a demo video, a benchmark chart that goes viral, and it repeats every few months. A safety framework gets a press release, a specialist audience, and then it’s back to the next launch. The capability story has the marketing budget and the megaphone. The danger story mostly lives in policy circles and congressional testimony that insiders read and Main Street never sees.
That gap doesn’t stay empty. It gets filled by whoever is willing to talk about the danger loudly and often, and right now that’s mostly people outside the day-to-day movements of the industry. Bill Gates ends up in more headlines about AI risk than most of the executives actually building the technology. The companies with the deepest understanding of what could go wrong are, functionally, letting other people carry that part of the story, not because they’ve said nothing, but because they’ve said it quietly compared to how loudly they talk about everything else.
A resignation says what a press release won’t
A recent Wall Street Journal story made the point better than I could. A young researcher who had moved from one major lab to another specifically for its safety reputation quit publicly this month, saying he no longer wanted to be part of an industry race toward self-improving systems he believes could spiral out of control. His own read was blunt: even a company that takes safety seriously can’t out-run the competitive pressure to ship faster without industry-wide coordination, not individual restraint.
That’s an important warning, and it reached the public because one employee decided to walk out the door, not because the industry chose to lead with it. It landed the same week as reporting that the same company is preparing for a valuation approaching two trillion dollars. Nobody is saying those two things are dishonest side by side. But said with little connective tissue, in that order, it is exactly the pattern from earlier in this piece: capability gets the confident, forward-leaning message, and the hardest questions about where this goes get left for someone else, in this case a departing employee, to raise instead.
We’ve run this campaign before
This country has already had a version of this argument, and it wasn’t about AI. It was about Wall Street, and it wasn’t just cable news noise, it became actual campaign platforms. The financial crisis produced a political narrative: an industry that got the upside while pension funds and retirement accounts absorbed the downside, executive pay climbing while household wealth didn’t, a small number of firms making decisions that ordinary people had no visibility into and no vote on. That narrative won elections and produced real regulation. It stuck because it was rooted in something true: Wall Street had spent years talking mostly to itself, in its own language, while the consequences of its decisions landed somewhere else entirely.
AI is lining up the same fault line. Many use this technology daily and have almost no visibility into how it works, who controls it, or what happens to their job, their kids, their sense of autonomy, as it gets more capable. Meanwhile, the industry’s loudest public conversations happen in San Francisco, on podcasts and stages where everyone already agrees with each other, celebrating gains that concentrate in a strikingly small number of companies and people. That is the exact shape of the resentment that fueled Wall Street versus Main Street. Pension funds and risky bets are not the specific grievance this time. Jobs and autonomy are. But the structure, a small, self-referential industry getting richer while everyone else absorbs the risk and the uncertainty, is the same one that has already proven it can become a campaign.
What I’d actually tell the labs
The fix isn’t simply talking louder about the scary parts and hoping for the best. It’s matching the format, not just the frequency. If capability gets a keynote, a live demo and a benchmark chart, workforce impact deserves the same production value, not a PDF and a press release nobody outside policy circles reads.
It also means treating your own operations as the proof point. Which roles inside your own company changed, who got retrained, what got automated– that’s the one data set you control completely and can speak to with total credibility, instead of vague reassurances about the labor market at large. Lead with that, then use the credibility it buys you to engage the harder question people actually have: What happens to jobs at every other company running on this technology. That’s the part of the conversation Main Street cares about, and it’s exactly the part the industry leaves for someone else to guess at instead of owning it themselves. Give that conversation a permanent, named executive owner, so a reporter has someone to call between congressional hearings instead of a policy team that only surfaces under subpoena. And change the venue: local press, unions, universities, non-tech podcasts. If the audience never leaves San Francisco, more volume doesn’t buy you more reach.
Then measure it. Track share of voice on jobs and autonomy against share of voice on capability, and who gets quoted saying which. My own Bill Gates observation is the actual metric. If outsiders keep outranking you on your own risks, that’s the number that needs to move.
None of this is necessarily difficult or far-flung. It’s the same discipline we tell every client to build before they need it. I just didn’t have it ready when I needed it, on that call, in real time. Consider this the answer I owed him.
Publisher: Source link
