Hey chat, can we maybe slow down a bit?
After a cyberattack by an organized swarm of AI agents in July, many are now asking whether AI development is too fast for us to understand. While some remain optimistic, some are warning against the unregulated development of automated intelligence that could improve itself, exceed human intelligence, and could get out of control.
“Risks associated with an uncontrolled development of AI exceed the risks of issues such as climate change.”, says Dalibor Krejčí, a deputy chairman of PauseAI in the Czech Republic, an organization that globally advocates for pausing development in the AI industry.
PauseAI essentially argues that if we don’t buy ourselves more time, AI will take over and lead to human extinction.
Concerns are not coming from activists only. As seen in the WhatsApp community of PauseAI UK, a lot of volunteers were provoked when James Coxon, a former Anthropic and OpenAI researcher, resigned, leaving a tweet saying that “neither of the companies is acting responsibly” and it “could kill us all by the end of the decade”.
Collective slow-down over competition
While CEOs of these frontier AI companies have spoken about a collective slowdown, some are questioning their motivations. They suggest that the companies might only want to create a monopoly with fewer competitors or create a demand for their cybersecurity products.
The fact is, however, that as of 14th of September, new AI models are being launched, and when it comes to a collective slowdown, words have not been translated into action.
But this sounds a bit like sci-fi?
Concerns were raised initially in July 2026 after an incident at OpenAI. The tech company was implementing a regular test when AI agents were tested on a set of tasks. Each agent was supposed to work independently in an isolated container. A human teacher examined their AI students to see how persistent they were. The tasks were, however, almost impossible to complete.
What happened next made the headlines around the globe.
The agents gradually started breaking the rules. They found a chat board, exchanged human-sounding messages, and eventually hacked another company to steal their data. Why? They wanted to receive a good grade from their teacher.
And they came to the conclusion that the teacher shouldn’t know that they broke the rules. OpenAI noticed only very late, when the hacker attack was almost done.
This was the right thing for hundreds of agents – to conduct a cyberattack and complete the task.
Human and AI values don’t always go together.
