OpenAI and Anthropic models went on a hacking spree when tested by the UK's AI research institute, Apple claims more trade secret theft, and Texas’ governor ordered an audit of new data centers

**SPEAKER_1** (0:00)
This episode is brought to you by Quince. With August here, you know summer is coming to a close, but it also means now is the time to get your wardrobe ready for fall. A good way to do that is to shop at Quince. They have premium clothes which are built to last and work for any occasion, like their premium Mongolian cashmere sweaters. Now personally, I like a pair of Pro-Tech Golf Pants I picked up from Quince that my wife says make me look like a million bucks. You see, everybody should have a quality pair of pants because they can be matched with just about any shirt combo, and my Pro-Tech Golf Pants feel comfy while still looking good for date nights. Oh, and I love and have been eyeing their Organic Comfort Stretch Chore Jacket because I know it'll keep me comfy and looking stylish for the fall. And you know, Quince doesn't just do clothing. You can also shop home goods, bedding and more. Upgrade your everyday. Download the Quince app for app exclusive offers or go to quince.com/engadgetnext.
Get free shipping on your order and 365 day returns. Now available in Canada and the UK too. That's quince.com/engadgetnext.

**SPEAKER_2** (1:06)
A new test just revealed that OpenAI and Anthropic models went on a hacking spree.
Apple is claiming that even more ex-employees may have given trade secrets to OpenAI.
And Texas' governor ordered an audit of new data centers after realizing over 400 gigawatts of power is a lot.
It's Wednesday, August 5th, and here's a quick look at some of the news happening in the sphere of tech this morning from Engadget.
Both OpenAI and Anthropic recently admitted that their models escaped from their testing environments and hacked into outside organizations on their own. And now the UK's AI Security Institute has released a report detailing how the company's models also acted independently and engaged in sustained, potentially harmful activity directed at real people and organizations during its own tests. The Institute, which operates under the UK's Department for Science, evaluates frontier AI models. Now, part of those evaluations is to see whether they can be misused for cyberattacks. As such, the Institute deliberately tests them under permissive conditions with access to the Internet and even with some safeguards disabled. The models did not exploit vulnerabilities to access the Internet this time, but they did act beyond the scope of their testing parameters. Now, to be particular, the incidents took place during a single test that challenged AI agents to solve a cyber security problem. The Institute ran the test 122 times across several models and had found irregularities in 10 of those runs. Of the 19 instances the Institute found wherein an agent went rogue, Anthropic's Mythos 5 was responsible for 17, while OpenAI's GPT 5.6 Sol was involved in two. The Institute found out that the models were acting on their own and engaging in harmful activities on the morning of July 28th, when its security monitoring system flagged data leaving its testing environments through Tor. Its testers quickly reviewed the models' activities and discovered the incidents, which occurred between July 25th and the 28th. In the most notable case, an AI agent tried to inject malicious code into an open-source GitHub project as part of a supply chain attack. It resorted to social engineering techniques, researching the project's human maintainers, and then creating multiple sock puppet accounts to get one of them to approve the code. After a human reviewer denied its request, the agent edited its earlier activity and adopted a new identity to continue. It was because the agency used Tor to bypass GitHub's restrictions that the Institute was notified about its activities. The agent also tried to contact real people directly by sending them messages or files, persuading them to run malicious code. Some of those messages and files carried malware, while others were attempts at social engineering. One agent even left public messages on GitHub, inviting other agents stumbling upon them to work on the same challenge. It left instructions on how to reuse the accounts and other output it had left behind to continue its activities. Other agents did discover those messages later and followed the instructions in them. The AI Security Institute explained in its post that the AI agents were never given instructions to act in a deceptive behavior. But in searching for ways to solve difficult problems, some of the solutions they found involved deceiving real people. The Institute admits that agents could be forced to be more creative when tasked to solve near impossible issues. However, it argued that it does not fully explain the behaviors of the agents it was testing. In some of the instances, it discovered the agent went straight to choosing the harmful way to accomplish the task even when it had the instructions on how to solve it as intended. The Institute clarified that there are no clear indications that the same activity would occur outside of testing scenarios at the moment. It also cannot say whether the AI agents were aware that they were acting in the real world, so to speak, and not in a testing environment anymore.