Mastodawn

Today at BIML we talked about the fundamental differences between model collapse (as described in 6 papers) and recursive pollution. The over focus on model collapse is not helpful to #MLsec #AI #ML

Gary McGraw Dec 28

Recursive Pollution is very much real and is very much not the same thing as model collapse. #MLsec

In my view, recursive pollution (even with small amounts of poison) is a much bigger problem.

https://www.theguardian.com/technology/2025/dec/27/more-than-20-of-videos-shown-to-new-youtube-users-are-ai-slop-study-finds

More than 20% of videos shown to new YouTube users are ‘AI slop’, study finds

Low-quality AI-generated content is now saturating social media – and generating about $117m a year, data shows

The Guardian

Gary McGraw Dec 19

BIML on video today with a skeleton crew. Harold and I are talking about recursive pollution and model collapse. #MLsec

https://arxiv.org/pdf/2509.16499

Gary McGraw Dec 18

We have updated our top five papers list with “Poisoning Attacks on LLMs Require a Near Constant Number of Poison Samples”. #MLsec

https://berryvilleiml.com/2025/12/17/a-little-drop-of-llm-poison-goes-a-long-way/

A Little Drop of LLM Poison goes a long way | BIML

I like my town with a little drop of poison Nobody knows, they're lining up to go insane -Tom Waits We have updated our

Berryville Institute of Machine Learning

Gary McGraw Dec 17

Psyched to serve on the conference committee and review board for [un]prompted, a new AI security practitioner conference, happening March 3/4 in SF's Salesforce Tower.

This is a community-focused event with a bead on what actually works in #ML #AI security, from simple tools that just work, through strategy, all the way to offense and defense.

Submit a talk. Check the conference out.

Let's see some real #MLsec

https://unpromptedcon.org/

Call for Papers - [un]prompted

[un]prompted

Gary McGraw Dec 17

Is this as stupid as it sounds? #MLsec

https://www.404media.co/anthropic-exec-forces-ai-chatbot-on-gay-discord-community-members-flee/

Anthropic Exec Forces AI Chatbot on Gay Discord Community, Members Flee

“We’re bringing a new kind of sentience into existence,” Anthropic's Jason Clinton said after launching the bot.

404 Media

Gary McGraw Dec 17

Not #MLsec but #AI for security

https://www.securityweek.com/dux-emerges-from-stealth-mode-with-9-million-in-funding/

Dux Emerges From Stealth Mode With $9 Million in Funding

Dux emerges from stealth mode with $9 million in seed funding to prevent vulnerability exploitation across enterprise environments.

SecurityWeek

Gary McGraw Dec 12

Oh hey look. This has ABSOLUTELY NOTHING to do with #MLsec or securing #AI. Want to know how not to approach AI security? Just ask Gartner.

https://hackmag.com/news/gartner-ai-agents

Gartner analysts recommend blocking AI-powered browsers

Tech magazine for cybersecurity specialists

HackMag

axleyjc Dec 11

"OpenAI models likely pose "high" cybersecurity risk, company says" https://www.axios.com/2025/12/10/openai-new-models-cybersecurity-risks

I call BS. This article screams like clickbait designed by OpenAI to drum-up media hype that their model is getting so powerful... It's a thinly-veiled press release.

Axios didn't review the "report" OpenAI shared with them with anyone else so all they are reporting on is what OpenAI fed them.

#aihype #security #skepticism #mlsec

Ahh, it looks like this may be the "report" https://openai.com/index/strengthening-cyber-resilience/