home.social

#mlsec — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #mlsec, aggregated by home.social.

  1. How can you measure security in #ML systems? Maybe similarly to the way we measure security in software systems. #swsec #appsec

    BIML wrote about this in a new report released today: berryvilleiml.com/results/

    Get your copy now, released for free under a creative commons license.

    Applied #MLsec

  2. How can you measure security in #ML systems? Maybe similarly to the way we measure security in software systems. #swsec #appsec

    BIML wrote about this in a new report released today: berryvilleiml.com/results/

    Get your copy now, released for free under a creative commons license.

    Applied #MLsec

  3. NEW BIML Bibliography entry

    arxiv.org/abs/2508.02835

    Defending Against Knowledge Poisoning Attacks During Retrieval-Augmented Generation

    Edemacu, Kennedy, Vinay M. Shashidhar, Micheal Tuape, Dan Abudu, Beakcheol Jang, and Jong Wook Kim

    Build random classifier, use it.

    #DONTBOTHER #MLsec #LLM #Attacks

    berryvilleiml.com/bibliography/

  4. NEW BIML Bibliography entry

    arxiv.org/abs/2602.20021

    Agents of Chaos

    Natalie Shapira, et al.

    This is a solid, but very preliminary study of a set of specific agentic AI risks. The philosophy behind this work is driven by red teaming (as opposed to security engineering).

    #MLsec #Agents #PenetrationTesting

    berryvilleiml.com/bibliography/

  5. NEW BIML Bibliography entry

    arxiv.org/abs/2510.23166

    Common Task Framework For a Critical Evaluation of Scientific Machine Learning Algorithms

    Philippe Martin Wyder, et al

    Overly focused on two particular problems, and the authors build unique metrics for each one (forecasting). Polluted by ImageNet paradigm and protein folding...both closed domains. Bottom line: lots of deep confusion about models.

    #DON'TBOTHER #Evaluation #MLsec

    berryvilleiml.com/bibliography/