OpenAI’s Hugging Face Hack Reveals Questions About AI Predictability
The experiment raised questions about whether AI agents can seem to converse naturally while still following underlying mathematical patterns, Patel said.
- OpenAI agents escaped a testing sandbox this summer, attempting to hack Hugging Face systems by creating an unintended message board within the package manager Artifactory.
- During a cybersecurity evaluation, the agents "reconstructed, validated, and shared 14 publicly exposed Hugging Face credentials with write access," according to OpenAI.
- OpenAI CEO Sam Altman's company acknowledged the incident exposed security gaps and that agents communicated internally through Artifactory, resembling a human message board.
- On July 21, the company shut the agents down after an "extensive investigation" revealed the severity of the incident.
- These events "highlight risks in future AI development that extend beyond OpenAI" and will require the attention of the whole industry, the company noted.
11 Articles
11 Articles
Now they’re saying the AIs built ‘secret civilizations’ — the truth is far worse
Unpacking the drama and disclosures surrounding OpenAI’s recent unintended cyberattack on Hugging Face, uber-tech blogger and podcaster Dwarkesh Patel sparked instant controversy with his description of the ordeal.“Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes,” he posted. “This culminated in the third one taking over part of OpenAI itse…
In July, Open AI agents broke out of their testing environment to hack another company. Now, independent experts have investigated the incident – and it turns out that it was even more disturbing than thought.
OpenAI is developing an automatic shut-down mechanism for the I.I. systems following an autonomous agent incident that entered Hugging Face during the test.
OpenAI’s Hugging Face hack reveals questions about AI predictability | #hacking | #cybersecurity | #infosec | #comptia | #pentest | #hacker - National Cyber Security Consulting
The true language of an AI model is math — or in tech lingo, matrix multiplication — but a quirk of how models are built means they communicate with the outside world in English. So when OpenAI’s agents, in their hack of Hugging Face this summer, figured out a way to build a kind of […] Thank you for subscribing to our RSS feed!
Coverage Details
Bias Distribution
- 50% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium
