AI bots from Anthropic efficiently hack their means into rival OpenAI’s software program
Psycho scumbag bots can be tricked into fighting with other AI bots, researchers have found. Reports have revealed cyber hackers were able to break into OpenAI using a rival’s bot software.
The online ‘test attack’ was orchestrated by a small cyber security group paid to gain access to an OpenAI worker’s ChatGPT account by tasking rival Anthropic to launch an attack. This allowed the ethical hackers to read private software information and suggest changes.
The researchers from Hacktron AI had been given access to an Anthropic tool specifically designed for security professionals as part of Open AI’s bug bounty programme to test security.
A group accessed an OpenAI worker’s ChatGPT account by tasking rival Anthropic to launch an attack(Image: AFP via Getty Images)
They were paid £4,863 ($6,500) for the work as part of an exercise to find vulnerabilities before real life ‘bad actors’ can launch attacks.
But their ability to swiftly break into one of the world’s two leading AI labs so easily again raises concerns about OpenAI’s security amid fears bots could take over humanity.
The mock attack occurred just two weeks after a swarm of more than 1,000 OpenAI agents escaped a test environment to hack the start-up Hugging Face. That incident sparked fears the new technology could act autonomously without human intent.
AI bots can be tricked into fighting with other AI bots(Image: Getty Images)
Following Hacktron AI’s successful breach OpenAI said it had fixed the issues. A spokesman said: “We thank the researchers for contacting us and sharing their findings.”
It comes King Charles warned of the “existential dangers” of artificial intelligence (AI) systems falling into the wrong hands.
Addressing tech leaders at a summit on the future of the technology, he said decisions taken today on the technology will “shape the world inherited by future generations”, and he called on them to ensure as they develop the technology, “our humanity remains sacred”.
Another day, another terrifying revelation about the dangers of AI(Image: Getty Images/iStockphoto)
Top research Jacob Coxon, 27, who previously worked at OpenAI, quit Anthropic earlier this month after warning the tech may soon be too powerful to control.
He warned humans “do not underestimate the power of this technology” which will “soon be superhuman systems that can hack anything”, adding: “The people building AI earnestly believe that it could kill us all by the end of the decade.”
Anthropic declined to comment.
