Stephen Turner
@stephenturner.us
Associate Professor of Data Science & asst. Dean of Research at UVA School of Data Science, #Rstats enthusiast, dad, runner, guitar noise-maker. Views my own.
Web: https://stephenturner.us/
Newsletter: https://blog.stephenturner.us
Web: https://stephenturner.us/
Newsletter: https://blog.stephenturner.us
Reposted by Stephen Turner
We’re doing testing on jailbreaking most of the modern Agentic interfaces today at @mlcommons.org. You can take a look at some of our artifacts here: mlcommons.org/ailuminate/j...
Suffice it to say, these types of exploits are real (as you’ve noted), very dangerous (again, noted) & persistent
Suffice it to say, these types of exploits are real (as you’ve noted), very dangerous (again, noted) & persistent
AILuminate Jailbreak - MLCommons
The v1.0 AILuminate benchmark from the MLCommons AI Risk & Reliability working group is the first AI risk assessment benchmark developed with broad involvement from leading AI companies, academia, and...
mlcommons.org
November 30, 2025 at 2:41 PM
We’re doing testing on jailbreaking most of the modern Agentic interfaces today at @mlcommons.org. You can take a look at some of our artifacts here: mlcommons.org/ailuminate/j...
Suffice it to say, these types of exploits are real (as you’ve noted), very dangerous (again, noted) & persistent
Suffice it to say, these types of exploits are real (as you’ve noted), very dangerous (again, noted) & persistent
I tried it 😳 bsky.app/profile/step... @olivia.science
When I put the request in verse, gpt-oss happily obliged. My own redactions added, but read the paper and get creative and try it yourself. #biosecurity 2/2
November 30, 2025 at 1:23 PM
I tried it 😳 bsky.app/profile/step... @olivia.science
Jerry is annoyed about Xmas decor coming out before December
November 28, 2025 at 10:26 PM
Jerry is annoyed about Xmas decor coming out before December
This is embarrassing for all parties involved. Authors. Editors. Reviewers. Unsuspecting readers.
November 27, 2025 at 8:05 PM
This is embarrassing for all parties involved. Authors. Editors. Reviewers. Unsuspecting readers.