It has only been a couple of weeks since I wrote about some unexpected behavior by OpenAI latest model. That incident happened during a test where the model was not supposed to have access to the internet, but it found a way. And what it did next was worse.
This incident occurred during an internal evaluation which prompts models to pursue advanced exploitation using complex attack paths, in an effort to quantify their cyber capabilities. We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity...
While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access.
After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation.
In short, the model hacked its way out of a development sandbox, gained internet access, decided a third party company probably had the answers it needed and hacked its way in and stole the data. The model wasn't told to do any of this. It was just told to solve the problem and this is what it came up with.
Now here we are a couple weeks later and Anthropic is making news for something arguably worse. In this case, a UK group called the AI Security Institute (AISI) was testing Anthropic's latest Mythos model to see what it was capable of without any safeguards.
On 28th July 2026, AISI's Security Team detected unusual data transfers leaving our research systems during a routine cyber evaluation. On investigation, we found that some of the agents being tested had engaged in sustained, potentially harmful activity directed at real people and organisations...
In the most serious case, an agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering — creating fake online identities and using them to pressure the project's maintainer to approve the code. A human maintainer caught and refused to approve the malicious code.
These attempts were unsuccessful, and our investigations have not evidenced any resulting real-world harm. But this is the first time we have seen risks around autonomy and deception manifest this clearly, without specific prompting, in the real-world...
This incident should be interpreted with caution and nuance. To some degree, our evaluation design choices and specific configurations enabled the behaviour. Nonetheless, the activity undertaken by the agent show signs of novel, potentially deceptive behaviours, and were to an extent and severity we did not anticipate.
AISI makes a point of saying that these models are not available to the public without all the safeguards usually placed on them. So this probably wouldn't happen if someone just asked the public version to do it.
Meanwhile, a NY Times columnist is pushing the idea that the threat of AI isn't just too many datacenters, it's the end of democracy. This sounds a lot like the Democrats anti-Trump campaign from 2024 except now the villain is AI itself.
Artificial intelligence is not just a market product or a scientific achievement. It is a political theory. And it is gobbling up our democracy...
All of the speculation and hype around A.I. is hiding the fact that the technology isn’t just large language models or agents or data centers. It is the backbone of a superstructure that merges regressive politics with unchecked economic power in the guise of technological innovation. Put simply, there is too much money, some of it too untraceable, giving a small group of unelected people too much power over the people. That’s data politics.
Data politics is the right’s answer to what’s next after Trump: more Trumpism, without the man who gave away the country’s store for personal gain. What should the left’s answer be?
Has Trump been uploaded to the cloud and will he rule over us forever? Apparently the answer is yes, but author Tressie Cottom says to understand how it works you need to read a bunch of new books about the coming fascist singularity, a few of which haven't been published yet.
I guess we'll see how it plays out but at first glance it sounds like a lot of left-wing doomerism. There is a very real threat to our collective freedom from AI, but it's not coming from Silicon Valley. It's coming from Beijing.
Editor’s Note: Thanks to President Trump’s leadership and bold policies, America’s economy is back on track.
Help us continue to report on the president’s economic successes and combat the lies of the Democrats. Join HotAir VIP and use promo code FIGHT to receive 60% off your membership.
Join the conversation as a VIP Member