Cognitive warmup. Sundar Pichai, CEO of Alphabet and Google, went to deliver the 2026 Commencement Address at Stanford University. If I am to look beyond the excitement of the speech and the official blog post that followed with the speech in its entirety, turns out Pichai faced a mass student walkout and roundly booed during Stanford University’s 135th commencement ceremony at Stanford Stadium on June 14. The reason? AI.
In part, Pichai’s complete avoidance of even mentioning AI during his address is the biggest hint possible—the audience is no longer buying the pitch. Secondly, Google’s involvement in Project Nimbus doesn’t have as many admirers as the tech giant may have hoped. The Project Nimbus is a $1.2 billion cloud computing and artificial intelligence contract with the Israeli government and military, which is believed to enable surveillance and military operations.
“I know today is about giving you all advice. But people have also been giving me a lot of advice on what to say. Actually, it’s been the same advice, and it’s about what not to say. People thought it would be really difficult for me; it is the last two letters of my last name, after all,” Pichai said, in an attempt to diffuse the broader tension that is palpable.
CIRCULAR TATTLETALE
Something interesting happened, that defined days after Anthropic released the Fable 5 and Mythos 5 models last week, particularly calling the former “state-of-the-art on nearly all tested benchmarks of AI capability”, and noting the latter is “the same underlying model as Fable 5, but with the safeguards lifted in some areas”. That is when the party really started.
Amazon researchers reportedly found a way to bypass safeguards in Anthropic’s Fable 5 model. There are indications Amazon has findings showing the model could be used to obtain information relevant to cyberattacks. Amazon CEO Andy Jassy reportedly dialled senior figures at the White House, raising concerns about Anthropic’s new models. Soon after, Anthropic received a government directive to disable access to Fable 5 and Mythos 5—the US administration’s assessment seems to be that these two models have crossed some performance threshold that none other had thus far, and therefore caution is necessary.
Of course, Anthropic isn’t at best pleased. By the way, Amazon has invested roughly $8 billion in Anthropic and is its largest investor. Secondly, here’s the landscape— Amazon found a jailbreak, called the White House, got a competitor shut down, and protected its own products. The questions I am asking here are, were the same standards applied to every AI model that has claimed frontier capabilities till now (and there have been many; for the sake of funding and euphoria continuity) and have all frontier models that happen to be Fable 5 and Mythos 5 competitors, been tested using the same methodology?
Anthropic, while noting immediate compliance with the export control directive, does go on to say that “we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
VIBES AND HALLUCINATIONS
In October last year, one of the “big four”, KPMG published a report that is titled Total Experience: Redefining Excellence in the Age of Agentic AI, which writes realms about how companies are using AI to solve for customer demands and making the world a better place. Turns out, that was full of hallucinations. GPTZero, the maker of a popular AI content detection tool, have found inaccuracies and a reality of being littered fake footnotes all over the report.
GPTZero notes that only five citations out of 45 in the paper accurately pointed to real sources. As many as 28 citations paraphrased titles or added fake components to real sources, of which 12 were deliberately vaguely phrased to make it furthermore difficult to ascertain accuracy. The GPTZero researchers also called the creation of fake references by AI models as something called “vibe citing”.
Researchers also found that approximately half of the claims in the paper were fake or misattributed, and “likely the result of an AI research tool over-complying with a request to find examples of ‘agentic AI’ in the wild”.
I know this one very well, being an Emirates loyalist. KPMG claimed that Emirates has launched a mobile chatbot called Sara, powered by AI, which can chat with passengers and alter their flights. Turns out, Sara is a mobile assistant that was released on Emirates app and website in 2023, is not an AI-powered chatbot as it is claimed to be, and cannot change bookings for passengers.
Don’t hold it against me, if my trust on anything “AI” is a net zero.
GOT MY EYES ON YOU
If you have always wondered why Meta has historically been liberal with how it handles user data. Turns out, that trend is even more profound within the organisation. I read a report a few days ago that suggested Meta’s internal project called Model Capability Initiative (MCI) tracks employees across their work devices, in an attempt to get data to improve Meta’s AI models—that is, every mouse click and keystroke. Turns out, a very slim silver lining to this is that Meta will allow employees to “pause” this tracking for up to 30 minutes when they need to “check something personal”.
Recently, there was a leaked audio from an all-employee meeting in which CEO chatbot had defended the program saying AI can learn quickly from “watching really smart people do things”, and that the “average intelligence of people who are at this company is significantly higher” than the source of other data sets. Glimpses of a dystopian world that TV shows and movies liked to scare us with? They are the reality now.
