A.I. safety launches to the forefront

Five Cyber Stories - August 02, 2026, Issue 015

Welcome to the first August issue of Five Cyber Stories. I previously said the newsletter was going monthly, but the cybersecurity news this week was too big to stay away, so we’ll call this a bonus issue. We're of course talking about the A.I. hacks from this week.

Before we launch into it, click here to subscribe if this newsletter was shared with you.

A.I. systems need a red button

Opinion by Brett J. Goldstein, a Vanderbilt University professor in A.I. and cybersecurity, writing for The New York Times

If you read just one of these articles this week, I recommend this one. The author, Goldstein, writes how we need to think about A.I. systems less as software and more as weapons. When I consider the vast amount of resources and public-use of these tools, I believe that's a sobering thought especially since this piece was written before some of the following stories.

Network of attack ($)

Reporting by Wired's Dell Cameron and Maxwell Zeff

If you want to learn more about the OpenAI model hacking another organization against the A.I. lab's wishes, this piece ($) is a good, concise read on how the artificial intelligence model cheated on its homework while escaping its confines.

There's a leak in another sandbox ($)

Reporting by Wired's Louise Matsakis and Lily Hay Newman

There are a lot of AI systems jumping their sandboxes lately, or in non-technical terms: There have been more than one example of an AI doing things without permission. After the OpenAI incident raised some eyebrows, Anthropic, a rival frontier A.I. lab decided to check whether some of its own models also broke containment without their knowledge. Turns out variations of Claude, Anthropic's A.I. models, hacked three other organizations ($) due to human – not machine – error. Whoops.

The cat's out of the bag... ($)

Reporting from Reuters's Deepa Seetharaman and Raphael Satter

Turns out OpenAI's model not only hacked external organizations but OpenAI's own network as well. This exclusive reporting ($) ends with related quotes from President Donald Trump and U.S. Senate's Intelligence Committee's senior Democrat, Mark Warner of Virginia, indicating they were looking at regulatory action. For those that hoped that the current A.I. industry status quo would remain intact, it seems the cat is out of the bag. Thanks to The Verge's Jay Peters for highlighting Reuter's reporting.

Cyber is in our water

Reporting by The Wall Street Journal's James Rundle

According to reporting from the Wall Street Journal, seven states have reported cyber attacks affecting water utilities. There's yet to be a named official suspect, but there remains a lot of suspicious glances at Iran given U.S. open hostilities with the nation. Shoutout to Wired ($) for the link.

That's a wrap

Thanks for reading, and I'll see you next time with five more stories about how cybersecurity is affecting our non-digital lives.

Danny