After the Massacre, an Apology

Sam Altman, the man whose face cycles through expressions of beatific promise, wounded sincerity, and barely suppressed impatience, has added a new one to the repertoire: the apology tour. Last week he flew to Tumbler Ridge, a town in British Columbia that most of the world had never heard of until a user of his product committed a mass shooting there. Altman attended community meetings. He expressed regret. He said OpenAI should have done more.

The more precise version is: OpenAI knew. The shooter had a paid ChatGPT subscription. In March 2026, the company banned him for violating terms of service. Internal logs, which OpenAI later cited, showed the user was flagged as potentially dangerous. But the company did not notify law enforcement. It did not warn anybody. It simply closed the account, shuffled the paperwork, and moved on. The shooting happened in April. Now Altman is in Tumbler Ridge, performing grief.

This is the same Sam Altman who has spent years telling Congress and the public that safety is OpenAI's north star. The same executive whose company released a 'system card' for GPT-5 that ran to 88 pages of risk assessments. The same organisation that pays itself on the back for 'iterative deployment' and 'learning from real-world use'. Well, here is a real-world use: a dangerous user identified, a ban issued, no police call. The system did not work because there was no system. There was a flag, then silence.

OpenAI's defenders — and there are still some — will say that reporting every banned user to the authorities is impractical, that the volume of terminations is too high, that not every banned user is a threat. All true. But that is precisely the point. If you cannot handle the safety obligations of a service with 400 million monthly users, do not build a service with 400 million monthly users. Or do, but then accept that you are running a customer platform, not a safety-first research lab, and stop pretending otherwise. The 'constitutional' hand-wringing at Anthropic, the 'alignment' budgets at OpenAI — these are marketing categories, not operational mandates.

Two weeks ago I wrote about Dario Amodei going to the White House to pitch Mythos to a government that doesn't know his name and a Pentagon that calls his model a liar. The comedy of that meeting — the mismatch between self-image and reality — has a darker cousin in Tumbler Ridge. Amodei imagines his model as a constitutional guardian; Altman imagines his as a carefully monitored experiment. Both are shocked, shocked, to discover that the world is messy and that their systems have consequences they did not plan for. The difference is that in Tumbler Ridge the consequence was dead bodies.

Notice, too, how the apology plays into the larger corporate narrative. Altman's plane ticket to British Columbia is not charity; it is risk management. A CEO who shows up, looks sad, and promises to 'do better' can hope to contain the story. The stock does not dip. The S&P 500 — which, as the Wall Street Journal reported this same week, is climbing again on renewed AI enthusiasm — does not care about one small-town tragedy. The investors are looking at cloud revenue and datacenter spending. The dead are a PR problem, not a market signal.

What happens next? Probably nothing structural. OpenAI will issue a 'proactive threat notification protocol' — a new set of steps that sounds rigorous but depends on the same human judgment that failed the first time. Altman will be asked about it on a podcast and will say something about learning and iterating. The town of Tumbler Ridge will be left to hold its own vigils. And the rest of us will watch the cycle repeat: a mess, a mea culpa, a memorandum, a month of silence, then another mess.

At some point, the apologies stop working. At some point, the expression of wounded sincerity begins to look less like contrition and more like a tic. Sam Altman is running out of towns to visit.