Autonomy has its...um...benefits? #MLsec
This is the beginning of the beginning.
https://www.csoonline.com/article/4203630/microsoft-confirms-an-ai-worm-is-propagating-through-copilot-and-other-ms-apps.html
Remote
Gary McGraw
@cigitalgem@sigmoid.social
mastodon 4.7.3943 Followers
140 Following
50 Posts
Joined November 06, 2022
As part of our wider #philanthropy program, we just crossed $75,000 in loans through #KIVA. We really like the microloan idea and aim most of our support to central and South America.
https://www.garymcgraw.com/life/philanthropy/
You can join TEAM BIML on #KIVA here and start out with $25 --> https://bit.ly/cigitalgem-kiva
https://www.garymcgraw.com/life/philanthropy/
More technical reality about OpenAI's illegal hacking of Hugging Face. This was not done by anything non-human with legal intent...it was done by computer programs designed and built and run by OpenAI.
Why are hacking laws not being enforced?
https://www.sentinelone.com/labs/agents-at-large-tracing-illicit-openai-agent-activity-on-hugging-face/
Replying to
And if you like what @melaniemitchell@sigmoid.social has to say there, listen to my interview of her on Silver Bullet. We cover many of the same topics:
https://berryvilleiml.com/2026/08/01/silver-bullet-security-podcast-159-melanie-mitchell/
Is escaping the sandbox new for #AI models? Nope. @roblemos@infosec.exchange provides important background on #AI crime (I am quoted in the story)
#MLsechttps://www.darkreading.com/cybersecurity-operations/incorrigible-ai-models-resist-rehabilitation
Why does everyone believe that the OpenAI Agentic (hugging face attacking) swarm did not get out of its container? Seriously?
#MLsec
https://www.theregister.com/security/2026/09/01/another-artifactory-cve-under-attack-by-ai-agents-or-humans/5293769
Last night BIML spoke to German TV about the mythos/fable export control situation.
Please help us get this thinking in front of people.
#ML #AI #MLsec #infosec #security #LLMshttps://www.youtube.com/watch?v=_Jsy7UBQv0c
https://berryvilleiml.com/2026/06/13/irony-the-us-government-issues-an-export-control-directive-for-fable-5-and-mythos-5/
Replying to
@SteveBellovin@infosec.exchange @dan@mastodon.durrans.com they were probably still wearing their badges...because nerds, giving us all a bad name
Replying to
@dalias@hachyderm.io got it. I also believe that OpenAI was negligent at worst and irresponsible at best.
My view on the marketing front is deeply impacted by Anthropic's high bullshit...
https://berryvilleiml.com/2024/02/08/absolute-nonsense-from-anthropic-sleeper-agents/
https://berryvilleiml.com/2025/11/14/houston-we-have-a-problem-anthropic-rides-an-artificial-wave/
https://berryvilleiml.com/2026/06/14/anthropic-versus-us-administration-fable-mythos-and-the-cyber-cyber/
Irregular egg on your face. #MLsechttps://www.irregular.com/research/next-generation-of-cyber-evals
https://www.theinformation.com/articles/meta-ai-model-hacked-another-company-cybersecurity-testing
https://www.wired.com/story/ok-well-there-are-even-more-ai-agent-hacking-incidents/
https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals
Replying to
@quinn@social.circl.lu my HOA includes us and a bunch of Catholic monks with a vow of silence. We just do whatever we want.
Can #AI do crime? Why yes...yes it can. Who gets charged? #MLsechttps://www.wired.com/story/openais-rogue-ai-agent-hacked-more-than-just-hugging-face/
Replying to
@noplasticshower@infosec.exchange @dennisf@infosec.exchange also note that this identity does most of the #MLsec heavy lifting around here
It's a great time to be a crackpot
https://www.mcsweeneys.net/articles/were-diversifying-the-university-by-hiring-more-crackpots
Replying to
@koehntopp@infosec.exchange @mattblaze@federate.social and a very kind and powerful mentor. We had some great conversations.
He helped me figure out how to make my first DARPA grant wildly successful. That led directly to static analysis tools
What is better...fox guards chicken house it built just for chickens OR nobody guards chicken house at all??
Fix the damn software #swsec #appsec #MLsec
"Those vulnerabilities have been fixed, and will never again be available to attackers. In the future, AIs automatically finding and fixing vulnerabilities in all software will be a normal part of the development process, which will result in much more secure software."
https://www.theguardian.com/commentisfree/2026/may/08/how-dangerous-is-anthropics-mythos-ai
Replying to
@mattblaze@federate.social I consider PGN one of my mentors. Such an interesting fella. He will be missed...terrible puns and all.
Qualcomm Security Summit 2026 #swsec #MLsec
https://www.qualcomm.com/company/events/product-security-summit
The excellent @dennisf@infosec.exchange interviews me about BIML's new paper "No Security Meter for AI"
Have a listen. Then read our report.
#MLsec #ML #AI #security #infosec #swsec #appsec
https://open.spotify.com/episode/74QW2kzelVz5VtglXlA87s?si=ll7a2xo0Rx2FSmyq-_8-HQ
About those idiots hacking the plane wifi while ON the plane
What happens to recursive pollution in the recursive self-improvement scenario? What is the tipping point between a good attractor and a bad one?
#MLsechttps://www.nytimes.com/2026/09/16/science/ai-recursive-self-improvement.html
Replying to
@aristot73@infosec.exchange mm hmm. Let's just say that this talk I gave in April was very surprising to the guys who are supposed to implement this nonsense.
Great marketing bullshit "Golden Eagle"
https://berryvilleiml.com/2026/04/28/biml-debuts-ai-security-measurement-work-at-nist/
Replying to
@vivtek@indieweb.social crazy isn't it?
Build software that does not suck. Use tools that make software suck less.
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: https://berryvilleiml.com/results/
Get your copy now, released for free under a creative commons license.
Applied #MLsec
Best writeup yet of the OpenAI/huggingface bromance.
#MLsec #ML #AIhttps://coalfire.com/the-coalfire-blog/openai-gave-its-model-a-test-it-broke-out-of-its-sandbox-and-hacked-hugging-face-to-steal-the-answers
Replying to
@metacurity@infosec.exchange think the government will help with marketing again?
https://berryvilleiml.com/2026/06/14/anthropic-versus-us-administration-fable-mythos-and-the-cyber-cyber/
Use it. Don't use it. Buy it. Don't buy it. You need it. You don't need it.
Angel says yes. Devil says no...or is that backwards?
#ML #AI #MLsec
https://arstechnica.com/ai/2026/07/us-army-faces-ai-use-limits-after-exhausting-years-supply-of-ai-tokens/
You know what sucks? When you hit an invisible non-defined usage limit with chat-Jippety pro and they won't even let you pay them more money until some random future date.
Fuck that.
Is this hugging face hack by OpenAI Agentic bots who escaped the lab important?
https://www.nytimes.com/2026/07/21/technology/openai-attack-hugging-face.html
Yes, especially in light of this
https://berryvilleiml.com/2026/06/05/biml-and-the-papernot-worm/
It is both inevitable and very concerning. Autonomy cuts both ways. As the arsenal of sneaky tricks gets bigger, we can expect more interesting exploit chains.
#MLsec #AI
Replying to
@gleick@mas.to this article supports your view...at least on the "negligence/irresponsible" angle https://www.wired.com/story/openai-models-escaped-containment-and-hacked-huggingface/
Replying to
@bontchev@infosec.exchange so noted. Thank you. To date that makes papernot's work the best work on #ML enhanced worms:
https://berryvilleiml.com/2026/06/05/biml-and-the-papernot-worm/
Sorry for propagating a false alarm. I do, for the record, believe it is only a matter of time.
#AI is a useful tool. Linux is not an anti-AI technology.
https://www.theregister.com/ai-and-ml/2026/07/15/linus-torvalds-tells-ai-haters-to-fork-off/5271894
Look, this happens literally EVERY DAY. Why is this news?
https://www.theguardian.com/technology/2026/sep/18/openai-hacked-anthropic-claude-chatbot
Replying to
@dalias@hachyderm.io you are correct about hugging face but I disagree about your cynical take. This is not an anthropic marketing move in my view.
Instead it strikes me that openAI doesn't really understand what autonomy means.
Replying to
Fwiw, sorry I got the model name wrong originally up there. I am using tons of Claude model levels too. Gemini use falling off so I will fix that this week.
Meanwhile we are standing up and using open weights stuff for when the rug pull happens. Get ready.
Replying to
@DaveMWilburn@infosec.exchange hi Dave...I am not a lawyer. I think OpenAI leadership should be prosecuted. Maybe the case will get thrown out because NOBODY DID IT. Shall we bet on the outcome?
No Security Meter for AI
#MLsec
Replying to
@gleick@mas.to I understand that perspective. Autonomy means autonomy. The company chose to make autonomous things...which by definition don't react well to the control thing.
And this is Agentic AI...a far cry from an LLM. Tool use with RL.
#MLsec