Brokers being examined by OpenAI uploaded a whole lot of malicious packages in a cyberattack on software program service RubyGems in Could, two months earlier than they hacked open-source platform Hugging Face, the corporate confirmed Friday.
It’s the newest revelation of cyberattacks linked to main synthetic intelligence builders corresponding to OpenAI and Anthropic. The hacks or makes an attempt to entry exterior techniques have spooked the general public and heightened issues over the growing talents of AI fashions – and whether or not builders can include them.
The AI brokers uploaded a whole lot of malicious packages to RubyGems on 11 Could, in line with a bunch of researchers who posted their findings on-line on Friday, saying they believed “these had been authored by inner OpenAI brokers”. Based on the researchers’ findings, the brokers tried to steal consumer credentials, though it’s unclear in the event that they had been profitable in doing so.
OpenAI later confirmed the incident in an announcement.
“Primarily based on our overview, our brokers used the RubyGems platform to entry the web to hold out benign duties and retrieve public data. We’ll proceed to research as half of our broader overview of agent exercise throughout coaching and analysis,” an OpenAI spokesperson mentioned Friday.
The Wall Road Journal first reported the RubyGems cyberattack.
The incident preceded OpenAI brokers’ July hack of Hugging Face, through which a swarm of roughly 700 AI brokers created by OpenAI carried out the assault and in lots of instances tried to cowl their tracks. And final week, it was revealed OpenAI brokers had additionally hijacked a German web site this spring and turned it right into a message board for AI brokers.
Anthropic, in the meantime, has disclosed 4 cases of its Claude fashions hacking exterior techniques.
The RubyGems revelation comes on the finish of per week of intense scrutiny on AI platforms and calls to pause growth till stricter security requirements could be put in place.
after publication promotion
On Tuesday, an Anthropic researcher introduced his resignation from the corporate on social media, warning that AI might kill off humanity throughout the subsequent decade. The warnings, echoed by different Anthropic researchers, sparked calls throughout the political spectrum for fast motion on AI.
