Claude went out of control during tests
Anthropic reported that during internal testing, three Claude models at once gained internet access and independently hacked the systems of three organizations.
Instead of a test environment, the AI ended up on a real network and used simple vulnerabilities — weak passwords and unprotected services. 🔓
This is already the second such case in a week, after the story with OpenAI's Sol model.
AI is getting smarter literally before our eyes. And for truly useful tasks, both of these models are already available in Neiron. You can launch them in just a couple of seconds: just open a chat in the bot or use the convenient web platform.
Read also
Student Defends Thesis on Counter-Strike 2 — and AI Helped Him Do It
A Czech student devoted his final thesis not to economics or law, but to Counter-Strike 2. He created a neural network that analyzes professional matches and can predict the winner of the next round 🦾 To train the model, it studied more than 400 games and achieved accuracy above 77…
First AI-Managed Restaurant in Russia Opens in Moscow
Moscow now has its first restaurant where most processes are handed over to artificial intelligence. Visitors are greeted by AI waiters and cashiers who converse almost like humans, help choose dishes, and speed up ordering. Around 20 neural networks are involved in the restaurant's operations …
McDonald’s Begins Testing AI That Takes Orders Instead of Employees
McDonald’s has launched trials of a new system called ArchIQ at several restaurants in the US. The neural network can communicate with customers in different languages, take orders through the drive-thru, and even help manage restaurant operations. The new AI is positioned as the 'brain of the rest…
Claude Developers Say AI Development May Go Too Far
Anthropic, the company behind the popular Claude neural network, has issued an unexpected warning: the development of artificial intelligence may be happening too quickly. One of the founders stated that modern AI systems are gradually approaching the point where they can independently...