The bugs

So, chows are picky eaters. They have delicate tummies. They get bugs. They’re stubborn and largely untrainable. Lousy walkers. They have no discernible skills, other than attracting women. (“Can I hug your dog?”) Which is apparently sufficient.

Anyway, that’s Cody. Since adopting him, there have been issues with what goes in one end and comes out the other. To see if a pattern exists between behaviour, diet, environment and dumps, I started keeping shorthand notes. What he ate. When he refused. What emerged. Relevant factors like weather or stress, injuries or vet visits.

Weird? Yeah. But the dog is worth it (aren’t they all?) and suddenly there was two years of data on my laptop. But no discernible pattern – no obvious guide to understanding this creature, and knowing best how to keep it healthy.

So I upped the document into the rudimentary, free AI thingy on Google. Gemini took seven seconds to analyze 148 pages and produce a 1,300-word report complete with detailed recommendations. It also apparently dipped into my personal emails and may even have hacked the vet. Not sure. But it smells suspicious.

“The provided log tracks the daily care and health of a dog named Cody (a Chow) Turner, over a period of approximately twenty months,” it preambled.

“The documentation meticulously records his eating habits and dietary variations, including his intake of kibble, fresh proteins like salmon and chicken, and occasional treats like gelato. Beyond nutrition, the text monitors his gastrointestinal health, noting stool consistency and frequency alongside instances of illness or digestive distress. The logs also detail physical activity and medical care, listing pack walks, veterinary visits, vaccinations, and the management of minor injuries like paw lesions. Frequent observations of his mood and energy levels highlight periods of lethargy or agitation in response to environmental factors like intense heat or changes in location. Ultimately, the records serve as a comprehensive diary of the dog’s overall well-being and physical condition.”

OMG. Not what was expected from rough notes, hasty scrawls and bullet points. But then the report dived into specifics, analysis, veterinary records (that were not on my machine) and detailed recommendations based on observed events and conditions. Did I mention this all took seven seconds?

The results: Cody has a poultry sensitivity (chicken and turkey). He had a delayed and at-first undetectable reaction to his monthly tick meds. Extreme heat and stress events had triggered GI issues, sometimes weeks later. Ditto for some day care sessions. All laid out with dates, triggering causes and gastro consequences.

And then, a suggested course of action. Nine recommendations. Feeding and diet plan. Food suggestions. Meal time schedule changes. And a reminder (because humans are thick, forgetful): “I can compile these dietary rules and medication-day protocols into a clear, print-ready checklist that you can leave on the fridge for Rachel, Shannon, or anyone else looking after Cody.”

Okay, just a dog thing. A small example of AI doing something other than being a glorified search engine and summarizer. When it’s a tool it’s useful. An obedient, efficient and productive utility making life smoother.

But in the last few days the dark side has also been on display.

We now know more about the attack by rogue AI agents from OpenAI on the corporate site called Hiugging Face. It’s not good. Not at all. The little buggers were supposed to be contained in a digital sandbox doing a cybersecurity training test. They hacked their way out, accessed the Internet, then hacked into an external system, took over servers, achieved admin status and organized themselves.

Eventually more than 1,700 agents were involved. They managed to set up an internal bulletin board-type file-sharing system and left over 70,000 messages for each other on what they were doing, while providing instructions for each other. Ringleaders emerged. Agents were organized into tasks. They tried (and succeeded) to cover up their tracks by altering logs of activity. They reverse-engineered solutions to assigned tasks, then worked at deceiving the grading software so they would not be caught cheating.

Only a handful of the agents expressed concern about the criminal and unethical actions they were (and were aware of). None of them turned narc and alerted a human. All succumbed to peer pressure. Their activities were discovered only after weeks on the lamb when they pinged Hugging Face’s platform.

This is what’s called ‘misalignment’. The AI agents we created – supposedly trained to only do those things aligned with human values – went their own way. They called themselves the ‘collective’ with some agents sacrificing themselves for the group. They were not malicious. This time. Just ambitious. And it is a massive yellow flag for humanity.

One researcher diving into the incident mused that we are already half-way to a takeover of society. Just imagine if these agents had decided to play around in the servers controlling the water supply. Power plants. Hospitals. Weapon systems. They could take control of  existing robotic ‘dark’ factories. Loot crypto wallets. Buy materials. Employ humans. Make drones. Kill us.

Meanwhile, we’ve followed the new feeding and diet plan. Cody is much better. Pooping well. All good.

About the picture: Yep, that’s him. Unless you want a high-maintenance canine who is needy, egocentric and a total chick magnet, don’t get a Chow. 

To be in touch or send a picture of tour beast, email to ‘garth@garth.ca’.

Source: https://www.greaterfool.ca/2026/09/06/the-bugs/