Claude Mythos Preview found new attacks on cryptography in a research-led test. The target was a weakened version of AES, not the standard running your bank or your messages, so nothing in use today is exposed. What holds the internet together is the assumption that certain math costs too much to break, and that assumption was priced for human attackers on human timelines. AI has moved faster in security than almost anywhere else. Often, the systems we trust most are the systems we have stopped examining.

"In the testing, Mythos was able to break the weaker version of Advanced Encryption Standard in a way that Anthropic said made an assault 200 to 1,000 times faster than what previous human research had managed to do."

I've felt this way for a long time. It's humorous when a company that has built it's business on scraping the web files a lawsuit to stop someone from scraping them. So many thoughts come to mind.

You reap what you sow.
What's good for the goose is good for the gander.
Live by the sword, die by the sword.
Rules for thee but not for me.
Turnabout is fair play.

The pendulum is going to swing the other way on this one. Panagram is jumping on the fear that exists around AI and the "AI Slop" framing of anything generated by AI. But the reality is that someone using AI to help them write is not the same thing as 100% AI generated and autonomously published content.

I use AI to assist with writing, research, ideation all the time. But when I hit publish, I stand behind the content. I've researched it. I've directed it. I'm not shy about that. But hey, let the witch hunts begin.

Ben Thompson argues the panic over Chinese open-weight models like Kimi K3 is largely overblown economically: running inference on any model costs money, and that cost is directly tied to revenue in a way R&D spend is not, so "free" open weights don't mean free to serve, and once compute constraints ease, intelligence increasingly behaves like a commodity where the labs with the best cost structure win.

If history is any judge, I'll succumb to shiny object syndrome long before my iPhone becomes obsolete. This is likely the case for most iPhone users, though I know more than a few people who have not updated their iPhones in my years. The biggest sign being the charging cable they use. USB-C was added to the iPhone 15, which launched in September of 2023, and one of my friends was still using what appeared to be an iPhone 13 last week. Apple is continuing support on the new OS all they way back to iPhone 11, so that bodes well for longevity of the new iPhones.

This article concludes by saying, "Recreating events that were never recorded is an intriguing use case for AI." And I have to admit, the recreation of an unrecorded Pelé goal is certainly compelling, but it all just feels like a slippery slope. When we can twist "recorded" history to our whim, we are begging for trouble.