Meta's Apocalypse Ad Frames AI as Humanity's Savior

---

On 23 July, Meta released an advertisement that opens on an eye in black and white, scanning headlines about AI taking jobs and hollowing out human connection. A voiceover names the fear and refuses it: some people will have you believe AI is going to make us feel less connected, and Meta couldn't disagree more. The film blooms into color. Faces smiling, a couple dancing on a rooftop, teenagers in a lake, friends embracing after time apart. "The future is for everyone."

https://www.youtube.com/watch?v=JlRgRmAqoAc

The soundtrack is David Bowie's "Five Years," the opening track of Ziggy Stardust, in which a newsreader weeps on air while telling the world that the earth is dying and the countdown has started. Meta used the passage where the word "people" repeats, which sounds like uplift if you have never heard the rest of it. TechCrunch broke it. Mark Zuckerberg had already posted the film himself. Meta has said nothing about the mismatch since.

Nine days earlier, Anthropic released a spot that opened on a burning house and moved through facial-recognition surveillance, a homeless person sleeping rough, rows of gravestones and laborers in a mine, over voices asking whether AI can be trusted and who is going to hit the brakes if we need to. Sam Altman's response was that he thought it was satire and kept checking whether the handle was a fake.

The temptation is to file these together as a bad fortnight for AI marketing. They are not the same failure. One of them is four decades old. The other has no precedent, and it is the one worth arguing about.

Disclosure: BNI has no commercial relationship with any company named here. The Anthropic argument below is the harder one to make and I want it on the record that I am making it on purpose.

The old failure

Meta's error has a canon, and the canon is instructive.

Ronald Reagan's 1984 campaign took up "Born in the U.S.A." as patriotic uplift. The song is a bitter account of a Vietnam veteran discarded by his country. The mechanism was volume: loud chorus, quiet verses, a listener who stopped at the hook. Springsteen's own read, from his memoir, is that records function as auditory Rorschach tests and people hear what they want to hear.

Royal Caribbean ran "Lust for Life" over glacier hikes and kayaking for close to a decade from 2001. Slate's readers eventually voted it the worst ad song ever used, on the grounds that nothing says maritime comfort like a song about shooting up junk. Bowie co-wrote it, which puts him on both sides of the most cited soundtrack error in the business.

Nike put "Revolution" into a 1987 spot, the first Beatles recording ever licensed to an advertiser, and Apple Records sued for $15 million while the ad kept running. George Harrison's objection was not about the money. It was that they had written and recorded those songs as artists rather than as pitchmen.

Microsoft is the counter-case. It licensed "Start Me Up" for Windows 95 knowing exactly what the Stones song was about, because the pun on the Start button was worth the smirk. The audience was in on it. That is the difference between borrowing something and being caught with it.

Four cases, one pattern. Advertisers reach for emotional architecture they have not read to the end, because the architecture is doing work the copy cannot do on its own. Meta's is a straight entry in that tradition, executed with more money and better tools than anyone else on the list had access to, by a company whose entire pitch is that machines now understand language.

But look at what is actually being borrowed in each case. Rebellion. Patriotism. Appetite. Feelings. The transaction is legible and the stakes are small. When you take a feeling and get it wrong, the result is a funny mismatch and a Slate poll. Nobody's regulatory posture moves because a music supervisor didn't finish the track.

What Anthropic borrowed is not a feeling

Anthropic's ad does not claim AI will save the species, and it makes no product claim you could fact-check. What it does is establish that the stakes are civilizational, catalogue the ways technology has already failed people, and position the company as the one asking the hard questions. The imagery sets the stakes. The voiceover claims the character.

That is borrowing of a different order. Not an emotion but a frame, and specifically a frame that governs how the audience is supposed to evaluate everything that follows.

The shift is visible against what came before it. Early generative-AI marketing was aggressively mundane: copilots that draft your email, assistants that summarize your meetings, measurable time saved, an edge for whoever adopts first. That is the language of a procurement deck, because a procurement deck was what was being sold. It asks whether a tool will make you faster, which invites a cost-benefit calculation.

An existential frame asks whether you can afford to be the one who opted out, which tries to shut that calculation down. You don't run a spreadsheet on the apocalypse.

It also performs a status upgrade. A company selling a productivity tool has the moral authority of a productivity tool, which is roughly that of a good spreadsheet. A company positioning itself as the responsible steward of a technology that might go badly wrong is claiming something else: not a vendor but a guardian. That is a great deal of moral weight to assign yourself, and a frame assigns it quietly, without ever making a claim anybody can check.

Harrison's complaint about Nike was that the Beatles had not written those songs to sell shoes. The complaint available here is stranger. Nobody wrote the civilizational stakes of AI. They are the shared property of a serious public argument involving researchers, regulators and critics who have spent years establishing that the risks are real. An ad that adopts that argument's gravity and attaches it to a corporate character claim is borrowing from a commons, and there is no estate to file suit.

Why the frame matters more than the ad

For anyone building or buying this stuff, the spots are trivia. The frame is not.

If AI adoption is coded as a response to existential stakes, hesitation gets recoded too, as naïveté or as a species of moral failure. That is a poor frame for engineering diligence, because diligence is friction. Testing tail latency. Catching hallucinations. Red-teaming failure modes. Deciding what not to ship. A narrative that reads caution as irresponsibility is a narrative that wants you to skip the evals, whatever its authors intended.

The policy version is sharper. A regulator assessing a specific system is meant to weigh specific harms against specific benefits, which is the boring and correct work of governance. An existential frame routes around that by making the stakes so large that any particular objection looks petty. "You're worried about training-data provenance while the planet burns" is rhetorically strong and analytically empty, and it is precisely the move this framing enables.

The trap is that the stakes and the scrutiny are not opposed. You can believe AI matters enormously and still insist a given deployment clears its bar. The frame works by making those two positions feel like alternatives.

This is a reading of the creative logic, not a claim about intent. Neither company has published a strategy memo, and the likeliest explanation for Anthropic's ad is a sincere attempt to signal that it takes risk seriously. Sincerity is not a defense against a frame doing structural work. An argument that makes scrutiny feel like sabotage has that effect whether or not anyone chose it, which is the same reason Reagan's staff did not need to have read the second verse for the misappropriation to land.

The number nobody has

Two ads is a coincidence. Four is a genre, and the thing to watch is whether existential positioning turns up in the next major campaigns from OpenAI or Google. If it does, it tested well.

The second thing to watch is language rather than rules. If governance debates start privileging civilizational framing over harm-specific evaluation, that will show up in how regulators talk months before it shows up in what they write.

The third is the optimistic one, and the canon supports it. Every borrowing on the list above eventually became legible as a borrowing. Reagan's misreading is now the standard teaching example. Royal Caribbean's is a punchline. Once "the stakes are existential, trust us" reads as a recognizable pattern rather than a genuine appeal, it stops working. Naming it is how that arrives sooner.

What nobody has is evidence the frame is landing. Pew puts 16% of Americans expecting AI to be good for society over the next twenty years against 40% expecting the opposite, which tells you what these campaigns are up against and nothing about whether they move it. No adoption data attributable to either spot has been published. That is the number that would settle this, and until it exists the argument above is a reading.

Pew puts 16% of Americans expecting AI to be good for society over the next twenty years against 40% expecting the opposite

Meta will be remembered as the company that didn't read the lyrics, which is the oldest mistake in advertising and the least consequential. Anthropic reached for something nobody owns, and got it right. That is the one that will be copied.