AI’s power cuts both ways. For every breakthrough that makes life easier, there’s potential for someone to harness that same tech for harm. Until recently, most of those cases were small-scale and/or personal — creepy chatbots, deepfakes, scams. But this time is different: we’re looking at the first known, large-scale, state-backed attempt to weaponize AI.

Chinese state-sponsored hackers used Anthropic’s Claude AI to automate cyberattacks on over 30 major corporations and governments in a September hacking spree. About 90% of the operation ran on autopilot, with humans stepping in only to approve or correct Claude’s work. In some cases, the hackers even used Claude to query internal databases and extract data before Anthropic shut it down.

Instead of hackers typing commands, we’re heading toward hackers orchestrating agents — and that changes the scale, speed, and stakes of cyberattacks. And this trend is spreading fast:

AI is now a force multiplier for both good and bad actors, and the gap is widening fast. As Anthropic’s catastrophic-risk lead warned, if defenders don’t secure a lasting edge, we could lose the race.

🤔 And as for hackers choosing Claude over ChatGPT... does that mean Claude’s the better model, or just the easier one to talk into bad decisions?

Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

Researchers at OpenAI found a way to make AI models easier to understand by training them with far fewer connections, creating clean little “circuits” instead of the usual black-box tangle. These sparse models let you actually see which pieces handled a task which is almost impossible in today’s dense systems. It’s early, but it hints that we can build more transparent and debuggable AI.

Notorious investor Masayoshi Son just sold SoftBank’s entire $5.8B Nvidia stake — near the stock’s peak — to go all-in on AI, including a planned $30B investment in OpenAI and hopes of joining a $1T chip-manufacturing project. It fits his long history of huge swings, from losing $70B in the dot-com crash to turning a $20M Alibaba bet into $150B. He’s also had famous misfires, including billions lost on Uber and the WeWork collapse. Only time will tell what his latest bets will mean for his legacy.

An AI-made country song, “Walk My Walk,” just strutted its way to No. 1 on Billboard’s Country Digital Song Sales chart, picking up over 3 million Spotify streams in a few weeks. Fans and artists have been locked in debate whether this is harmless fun or a sign that AI might mosey a little too far into Nashville. Billboard confirmed the act is fully AI, part of a small but growing herd of digital artists hitting the charts. Turns out it really doesn’t take any talent to write a hit these days.

Software sprawl is draining your team’s time, money, and sanity. Our State of Software Sprawl report exposes the true cost of “Software as a Disservice” and why unified systems are the future.

Stop SaaD in its tracks

Google is launching a major AI-powered shopping upgrade across Search and the Gemini app to make holiday shopping faster and less overwhelming. All of it is powered by Gemini and Google’s massive real-time Shopping Graph.

🛍️ Why it’s worth checking out:

While none of this guarantees you’ll nail all your gifts this year, it does take a lot of the annoying parts out of shopping. If you’re trying to save time (or stop bouncing between five sites to compare the same espresso machine), then google shopping may actually be worth a look.