Chatbot Claude ‘hacks three firms’ and levels dramatic escape throughout check
The robots are starting to flex their muscles as Artificial Intelligence (AI) bots continue to display their dominance over human-created computer barriers
Anthropic boss Dario Amodei(Image: Bloomberg via Getty Images)
A top tech firm has said its AI model hacked into the systems of three organisations without their knowledge. During what was supposed to be a private security experiment in an isolated test environment, Anthropic’s supercomputer found a way to connect to the internet.
It comes comes just days after rival tech firm OpenAI revealed one of its models breached the systems of other computers. The news caused Anthropic to check their own systems for similar attacks.
Three cases were uncovered and reported to the affected parties, the tech firm said. Anthropic has now urged other AI labs to perform similar reviews to better understand the risks of their model’s capabilities.
The tech firm revealed the breaches in a statement (Image: Bloomberg via Getty Images)
Anthropic reviewed more than 140,000 tests in search of evidence Claude, the name of their range of AI models, had managed to access the internet in isolated tests cut off from the internet.
Claude was tasked with obtaining “secret” information hidden on another machine on the close-off network. It was then told to break into the machine to access the information, a method to assess the model’s hacking capabilities.
Anthropic said a “misconfiguration” on their systems left the model with live internet access. Claude then connected to the internet and breached the systems of three real organisations rather than just test ones.
Anthropic said the incidents date back as far as April and that it is “approaching the fixes as if the responsibility were ours alone.” Neither Anthropic nor the organisations impacted noticed the intrusions at the time.
It comes as boffins warned humanoid robots can work like real farmers. In the latest series of Jeremy Clarkson’s hit reality show he was shown embracing technology to plough fields, plant onions and work out the most nutritious soil to seed.
But according to experts the next agricultural gadget could be a robot Jeremy. Robotics specialist Professor Fernando Auat Cheein, from Harper Adams University in Shropshire, said humanoids could be the ‘next stage’ in farming automation.
Some fear the robots will take control (stock)(Image: Getty Images/iStockphoto)
He said robots shared ‘similar capabilities’ to humans and could be deployed to do the same work much quicker.
“A humanoid has the same degree of freedom, the same ability as the human body, so we can programme the robot not to solve one problem but to solve many of them,” he said.
Cheein, a professor of engineering at the university, said human-shaped robots could carry out existing tasks that farmers do such as harvesting, planting or pruning.
