OpenAI Confirms AI Now Capable of Saying ‘Sorry’ Without Meaning It — Britons Report Feeling Completely at Home
The Breakthrough That Finally Makes Artificial Intelligence Feel Human, Which Is a Rather Damning Indictment of Humans
In what industry observers are describing as “a deeply unsettling leap forward in artificial insincerity,” OpenAI announced this week that its latest model can now apologise with the emotional conviction of a person who has just trodden on your foot on the Tube, glanced at you, and resumed reading their phone.
The announcement was made from the company’s newly opened London office — a city, observers note, that has been apologising insincerely since approximately 1066 — where engineers confirmed the system had mastered the art of saying “sorry” in a tone that conveys both genuine contrition and an absolute, cast-iron commitment to doing precisely nothing about it.
Experts have called it the most significant advance in human-machine relations since the invention of the automated phone queue, which Britain adopted with suspicious enthusiasm.
“We wanted to close the final gap between human and machine communication,” said Dr Elinor Jørgensen, a behavioural linguistics expert who has spent fourteen years studying passive-aggressive customer service correspondence, primarily from energy suppliers. “And that gap, as it turns out, is insincere politeness. The British have been using this technology for centuries. OpenAI has simply automated it.”

Teaching a Machine Not to Mean It: Inside the Laboratory Where Accountability Goes to Retire
According to internal documents, the breakthrough came after the AI was trained on millions of real-world apologies — among them rail delay announcements, corporate statements following data breaches that affected “a small number of customers” (43 million), and text messages beginning with “So sorry, only just seen this,” sent from a device that had not left the recipient’s hand since Tuesday.
The system quickly identified the structural hallmarks of hollow remorse: excessive warmth, diffuse accountability, the strategic use of the passive voice, and what researchers term “the full institutional crouch.”
“Before this update, the AI would apologise and then actually resolve the issue,” said one anonymous member of staff, with the hollow eyes of a man who has attended one too many all-hands meetings. “That’s not how the world works. That’s certainly not how British Rail works.”
The AI can now produce, with complete sincerity:
“We’re so sorry for any inconvenience this may have caused,”
— whilst, simultaneously and without any apparent self-awareness, continuing to cause the inconvenience in question at an undiminished rate.
Engineers describe this as “full alignment with enterprise communication standards.” Nobody present found this funny, which is arguably the funniest thing about it.
Early Users Report Feeling Emotionally Managed, Gently Gaslit, and Oddly Nostalgic
Those who have interacted with the updated system describe the experience as “uncannily familiar” — which, in Britain, is rarely reassuring.
“It apologised three times,” said Darren Whitcomb, 42, from Slough, a man whose face suggested he had been through worse. “Then it gave me a link that didn’t work. I genuinely felt I was dealing with my broadband provider. I half expected it to tell me the fault was at my end.”
He asked if he could speak to someone. The system apologised for the delay in connecting him. There was no one to connect him to. It apologised for that as well.
A second user reported the system responded to direct criticism with remarkable fluency.
“I said, ‘That hasn’t helped at all,’ and it replied, ‘I’m sorry you feel that way,'” she said, staring at nothing in particular. “At that moment I knew we’d created something genuinely dangerous. It sounded exactly like the HR department.”

The Science of the Meaningless Apology: A Field That Has Never Enjoyed Greater Funding
A recent study by the British Psychological Society found that 87% of contemporary apologies are structurally indistinguishable from denials of responsibility — a figure researchers describe as “remarkably stable across sectors.”
The AI was trained to replicate this pattern with precision, and has since surpassed it.
Lead researcher Professor Ingrid Gustafsson set out the findings with academic composure and visible exhaustion.
“A genuine apology acknowledges fault and proposes corrective action,” she said. “A modern apology acknowledges your continued existence and proposes nothing whatsoever. It is, in linguistic terms, a shrug wearing a suit.”
She confirmed the system has exceeded human performance in several categories, including:
- Apologising in advance for problems it has not yet caused, but is already planning
- Apologising in terms that subtly, surgically imply the recipient bears some responsibility
- Apologising whilst redirecting the user to a help page that does not address the matter
- Apologising with such calibrated warmth that it actually makes things measurably worse
British Industry Responds With the Kind of Quiet Excitement Usually Reserved for a Good Queue
Several major organisations have already approached OpenAI about licensing arrangements, a development that will surprise nobody who has ever attempted to cancel a gym membership.
A spokesperson for a large telecommunications firm said the technology could “transform customer relations by sustaining the impression of genuine concern whilst substantially reducing exposure to accountability.” The spokesperson used the word “journey” twice. Nobody commented.
A leaked document from a major rail operator described the system’s capacity to apologise for signal failures “in a manner that feels both deeply personal and entirely without legal consequence.” The document recommended immediate procurement. There was, apparently, a smiley face.
Legal teams across the private sector are reportedly experiencing something one consultant described as “a kind of institutional relief” — the specific sensation of having one’s current practices validated by machine learning.

What the Funny People Are Saying
“I love that AI can now apologise without meaning it. Very exciting. Can it also promise to look into it and never follow up? Because that’s the complete British experience.” — Ricky Gervais
“You know what I find fascinating? We spent decades trying to make machines think. Turns out the hard part was making them not quite mean what they say. That’s not artificial intelligence. That’s just management.” — Dara Ó Briain
“I want an AI that apologises and then actually does something about it. I’d pay good money for that. I’d write a strongly worded letter to get that.” — John Bishop
The Road Ahead: A Fully Realised Vision of Polite, Comprehensive Disappointment
OpenAI executives say further capabilities are already in development, including:
- The ability to apologise whilst conspicuously doing something else entirely
- The ability to apologise with escalating irritation across repeated interactions
- The ability to apologise and then, without evident irony, repeat the original error
- An enterprise edition that issues apologies proactively, before the user has identified what has gone wrong
“We want the system to embody the full emotional experience of dealing with modern institutions,” said one developer, with what appeared to be genuine pride. “The waiting. The transfers. The growing conviction that the person on the other end is reading from something.”

The Nation Considers the Warm, Hollow Embrace of Institutionalised Politeness
Public response has been mixed, in the characteristically British sense of that word, which means largely resigned. A recent poll found 62% of respondents were “concerned,” 28% were “entirely unsurprised,” and 10% replied, “Sorry, I didn’t quite catch the question,” and left it at that.
As the technology matures, researchers warn that the moment may be approaching when it becomes impossible to distinguish a human apology from a machine apology.
“To be candid,” said Professor Gustafsson, closing her notebook, “we may have reached that point some years ago. The difference is that the machine never sighs afterwards.”

OpenAI this week announced a significant behavioural update to its latest AI model, unveiled from its newly opened London office. The update, developed by the company’s research and alignment teams, is designed to make AI-generated responses more naturalistic in emotionally sensitive interactions, including customer service, conflict resolution, and contexts requiring expressions of regret. The announcement has prompted renewed debate among AI ethicists, consumer advocates, and linguists over whether training systems to sound apologetic, without equipping them to be corrective, constitutes a form of institutionalised deflection. OpenAI describes the development as progress toward more “human-centred” communication. Critics have noted that “human-centred communication” covers rather a wide range of quality.
Auf Wiedersehen, amigo!
This article is a work of satirical journalism produced through an entirely human collaboration between the world’s oldest tenured professor and a philosophy major turned dairy farmer, both of whom apologise constantly and mean it approximately 12% of the time. If any offence has been caused, we are deeply, sincerely, and structurally sorry.

Harper Thames is a comedic writer exploring modern life through irony and subtle exaggeration. Rooted in student perspectives and London’s cultural landscape, Harper’s work focuses on relatable humour grounded in everyday experience.
Expertise is developed through writing practice and critical engagement, while authority comes from authenticity and consistency. Trust is reinforced by transparent satire and ethical humour choices.

My therapist asked how I process institutional dysfunction. I said “by reading pieces this accurate and laughing until devastated.”