It's getting pretty old at this point, but Anthropic publicly announced a new version of Claude - Mythos - that was apparently too dangerous to release. Given the number of times OpenAI has played this card and the fact that Mythos was quickly largely found to be overblown, this feels like it is, again, only attention-seeking publicity. Or a distraction from the recent Claude Code leaks, who knows.
Well it was a great success... if you don't count the fake Doctors, fake patients, fake medicine, and overall fake promises. Too bad the facts paint the whole thing as more of a large-scale scam than an actual legitimate business.
As many others before him, NVIDIA's CEO Jensen Huang now lives in a totally different reality than most people, who do not mistake machines dumber than most 6-year-olds for actual intelligent beings.
It's obviously hard to say for sure, but rug-pulling Disney like this less than half a year after a "landmark agreement" isn't a great look. It feels like OpenAI was not being very earnest, but instead just doing anything it could to generate some hype.
More specifically, he said it's about "models capable of delivering product value for [...] millions of enterprises". While it's hard to point at an exact definition of "Superintelligence", one can point at many inventions that delivered product value for millions of enterprises without being remotely intelligent at all.
Let's believe that Anthropic's run-rate revenue was at least $1 billion in early 2025 and at least $5 billion in August 2025. Then let's assume they would calculate their run-rate in a straightforward way, say their last month of revenue times twelve. If both those things are true, then it should have made more than $10 billion in revenue in March of 2026, not even counting pre-2025 revenue. However, its Chief Financial Officer only feels comfortable stating the relatively low number of $5 billion when under oath. And while she did say "exceeding $5 billion to date", is anyone really to expect any company to downplay it's revenue by 50%? History kind of tells us that they tend to do the opposite, so what's going on here?
OpenAI went from prohibiting military use of its AI to abandoning its official commitment to safety to becoming a Pentagon contractor under Trump in just 2 years. Nice speedrun.
Jack Dorsey's Block announced that it would be laying off 40% of its staff because of "Intelligence tools" (so, AI). The true story is most probably a big restructuration after overhiring during Covid.
Although Anthropic was supposedly all about safety, to the point that that Dario Amodei alleges he refused to launch the company's first iteration of Claude a few months before ChatGPT to pursue internal testing, it is still dropping its Safety Pledge like it was no big deal. The pledge was to refuse to deploy an AI system that without guaranteeing it had proper safety guardrails in place. Guess safety is still less important than profit, after all.
A clip of Tom Cruise and Brad Pitt fighting went viral when it was claimed that it was fully AI-generated with only a "2 line prompt". However, the video turned out to be a face-swap on a previously recorded green-screen footage. Pretty cool, but still no AI magic.
It's probably safer to say that Cursor vibe-coded a prototype of a browser while relying on human labor to fix AI mistakes. Also the shared code was heavily criticized by the current Servo (another browser) maintainer, partly for "not implement[ing] web standards".
Sam Altman again makes bullshit claims, stating that "With GPT-5, now it's like talking to an expert, a legitimate PhD-level expert in anything, in any area you need". Too bad PHDs can't count the number of "r"'s in the word "strawberry".
Who thought that this would go unnoticed? Like, people know the actress who played the AI and just, recognized her?
Salesforce promised it needed 4000 less people to do the same work, but in doing so apparently outright lied, seeing how its leadership publicly backtracked on the claim.
In reality, the bank was casually trying to offshore its call center activity to India, where the call volumes increased so much that people who where fired had to be rehired and apologized to.
Not so long ago, Google proudly presented its net zero carbon commitment. Then came AI, power-hungry datacenters in toe. Then the commitment disappeared. Not so surprising, but still disappointing.
I mean, where to even start with this one? ChatGPT is a chatbot: it doesn't have any physical capabilities, it simply cannot even see or feel or understand the physical world. It can't even always count from 1 to 100 correctly, so let's simmer down here.
The AI industry could make some efforts to be actually transparent about its water usage, but instead we get this. It's bullshit and not even a real metric. Just because other people are making outlandish claims in the other direction doesn't mean you can get away with making up your own.
How the company was able to stay afloat for so long after having already been exposed in 2019 is beyond comprehension. But this time it's finally over for good.
Klarna was the company who's CEO claimed that AI can actually already do most if not all of our jobs. Which is probably why he is massively rehiring a year after, right?
Following media backlash and legal threats caused by its absolute disregard for copyright, OpenAI had promised to build a tool that would magically "identify copyrighted text, images, audio, and video" and offer a way to... opt-out of ChatGPT? This was so realistic that in the end it didn't happen at all.
Environmental pledges quietly shuffled away when an opportunity for profit arises? I mean, who could have seen it coming?
Remember "prompt engineers", the magical solution to the fear of AI cutting jobs? Well it might have never been a real position in the first place. If it sounds too good to be true...
In a truly shocking turn of events, the "AI" used by shopping app Nate to automatically buy stuff for you from anywhere on the internet was discovered to actually be humans operating from the Philippines. What a surprise!
Dario Amodei, CEO of Anthropic, claimed on march 10th, 2025 that AI was on the verge of more or less completely taking over software development (or at least code production). Turns out it was not even close.
In what was a lie so blatant some are still shocked to have read it, Sam "Just Trust Me" Altman declared on his blog last year that his team knew how to build Artificial General Intelligence, a kind of human-like intelligence but better. A year and a couple of months later however, he's apparently betting on "a new architecture" to save his AGI bet. So much for pure scaling!
OpenAI had early access to EpochAI's FrontierMath benchmark, although the latter was advertised as containing only "new and unpublished" problems as a way to avoid models already being trained on the data. But OpenAI's early access was not disclosed until after they released their last model at the time, which scored pretty well on said benchmark.
Wouldn't it be nice if running AI got cheaper so that anyone could, you know, actually turn a profit? Too bad the people claiming that it does only looked at one side of the picture.
Claims he was unaware of the extent of NDAs, despite his signature being on them. People rarely read legalese at length, but as CEO it's hard to believe he simply never knew.
Altman says "The voice of Sky is not Scarlett Johansson's, and it was never intended to resemble hers.", even though the voice is eerily similar to Johansson's AND Altman tweeted "her" at launch, hinting at the android character played by Johansson in the movie Her.
OpenAI has decided to dissolve its "Superalignment" (safety) team, less than a year after it publicly commited 20% of its computing power to it.
Although the word "could" is doing some heavy lifting here, we are now in 2026 and the last "Remote Labor Index" from Scale AI and the Center for AI Safety found that "The best-performing model achieves an automation rate of only 3.75%". This was, of course, comparing the AI's productions to actual human work. As of right now, Amodei's prediction still has six months to materialize, but it's safe to say that nobody believes it could happen.
Like too many others, Microsoft at some point built an AI that was supposed to be a kind of "gender detector" (although it doesn't and cannot work). It was supposed to retire it at some point, but apparently the engineers on the team never got the memo.
Robin AI, a legal tech company, claimed to be mostly or only powered by AI, which meant hiring an army of human labor to perform its legal reviews. The company eventually went belly up.