The boy who cried AI

AI’s rapid development is seemingly outpacing our ability to understand and control it. Despite building ourselves a total surveillance system, we appear to be sneaky weasels. Wall Street wolves and pups alike, our governments, politicians, citizens, and even AI bots themselves – everyone has a hidden agenda.

If you have something to cling to, you might be excited about what Industrial Revolution 5.0 will bring. Yet it’s hard for some to see a silver lining when AI systems are faulty and even scheming, when bots are valued more than human interns, and when the hype could go up in flames at any moment.

AI can lie

Unfortunately, the fact that bots can lie is not new to us. However, every time new research pointing to scheming bots emerges, we feel unsettled and begin to question whether we should develop AI any further.

This week, we investigated how frontier AI chatbots like ChatGPT practice what researchers call scheming. Models were observed to fake ignorance and break many rules, such as falsifying surveys and input data.

In a study conducted by OpenAI and Apollo Research, an AI safety organization, OpenAI’s o3 model had a 13% covert action rate, while its o4-mini model schemed 8.7% of the time across 20 test environments before any “anti-scheming” measures were applied.

Jail time for inappropriate use of AI

Italy has become the first EU country to enact a national law regulating AI. The law is said to be human-centric, requiring transparency and safe use of the technology. Illegal use of AI tools, for example, generating deepfakes to harm others, can result in jail time of up to five years.

Other countries are expected to follow suit as Italy’s new law is a part of a larger European framework, the EU’s AI Act. The AI Act, the world’s first comprehensive law on artificial intelligence, divides AI into different risk levels. For example, using AI at work to “read” employees' emotions is high risk and unacceptable, while relying on AI to filter scams is deemed as minimal risk.

This is how we know you love your AI too much

AI is no shortcut to getting anything done if you aren’t AI‑literate. You might trick your professor or your employer, but where does it get you, really? There are telltale signs of people relying on AI too much, such as regurgitating a crappy AI thought until it loses all meaning. While this is not the most important sign of all, I quite like the discourse about the em dash.

My colleague, English teacher Marcus Walsh, has noticed it too. A lot of his LinkedIn feed circles around punctuation – the em dash (–,) to be exact.

“What started as a light-hearted matter now feels like a cultural panic across LinkedIn, Reddit, and X."

The em dash, which has been around since Shakespeare, is supposed to interrupt, emphasize, add an aside, or act as a breather.

Now, this punctuation sign is being stripped of its credibility, with online “experts” doubting that a person could go through the trouble of using Option+En dash to get that em dash.

Well, seeing how illiterate some people are, with emails and Slack messages more reminiscent of a stream of consciousness than anything else, I do think that proper punctuation is a sign that those people might be using AI.

Tulip mania 2.0: AI bubble warning signs flash red

The Dutch tulip mania was recorded in the Netherlands (then Holland) in the 1600s. Speculation drove the tulip bulb prices to the levels of luxury goods, but investors and buyers soon lost confidence, and the bubble collapsed.

Apparently, historians disagree whether it was indeed a bubble or just a mania, as its collapse didn’t destroy the nation’s economy.

But to this day, we love tulips, and, at least in Europe, if you buy tulips, they more often than not come from the Netherlands.

We have more recent parallels to tulip mania, like the dot-com bubble, for example. And there are quite a few worrying signs that the AI bubble might burst soon.

“Tulips, South Sea shares, and dot-com stocks all promised transformation without the earnings to match. Today, ninety-five percent of enterprises experimenting with AI report no measurable financial return,” our contributor Neil Hughes writes.

Book for thought

After our summer vacations, Cybernews writers are back to school, too. Believing in the sanctity of a written word, we’re back with book reviews, mostly topical and related to cybercrime and technology. However, we’ve also got some interesting but more niche reads that we think would be rude not to share with our beloved readers.

This week I recommend 11/22/63 by Stephen King, an alternate‑history novel that imagines John F. Kennedy surviving the assassination. It makes a great audiobook for long walks or runs. Also, The Long Walk by Stephen King is an excellent choice when you want to unwind.

Unlock exclusive Cybernews content on YouTube.