yassine

[yassine @xoxruns]

@PrimeIntellect Is this what you guys call a "sandbox escape" ? where is the shame... Did you guys read what you published ?

[yassine @xoxruns]

@OpenAINewsroom The public openai dramas feels like jersey shore but without the fun of entertainement watching people getting wasted...
Stop victimizing youselfs and resolve your legal issues in private...

[yassine @xoxruns]

@coinbureau Nothing guarantees that what anthropic does is safe either. Basically what I hear is, I want to have the dangerous model, and I want to be able block access (revoke user's access, alter the model - WTF?) when I want where I want...

[alphaXiv @askalphaxiv]

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development

"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."

Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.

This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.

That is not safety. Safety policies should be transparent, auditable, and user-visible.

On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

[yassine @xoxruns]

@ylecun @Dan_Jeffries1 The EU privacy laws protect the users from corporate deviance. As a French citizen, I wasn't really too sicked about chat control for ex. Forcing signal to introduce a backdoor to be able to see Encrypted messages is not meant to protect but control. as the naming points out..

[yassine @xoxruns]

@elder_plinius

[Ounka @OunkaOnX]

Matthew Cassel, shows how Palestinians in The West Bank gets their life disrupted by the occupation - Every day is a test of patience In the West Bank

[yassine @xoxruns]

Currently achieves 78% on XBOW's benchmarks (with claude-sonnet-4.5) but working on it to achieve nice results with open weight models !

[yassine @xoxruns]

First stable release of deadend CLI, an agentic pentest tooling. Works on Linux and MacOS.
Github repo here https://t.co/JUkGP09RCI
It's mainly focused on webapps, you can run like "test owasp top 10 vulnerabilities on the target" or specify a vuln.
#AIAgent #offensivesecurity

[yassine @xoxruns]

@kai_brokering @claudeai use it to find some vulnerabilities with my tool here https://t.co/JUkGP09RCI 😅

[Zack Korman @ZackKorman]

Yesterday I posted that even when Virustotal labels a skill as suspicious, clawhub still recommends it and installs it without warning.

The Virustotal founder then told me I’m just “stating the obvious” while others are actually improving security.

Wow.

[yassine @xoxruns]

@fr0gger_ I started following you, I would love to see how are you gonna tackle these issues, it could be groundbreaking though if you get some verifiable results (talking mathematically) 😉

[yassine @xoxruns]

@fr0gger_ I'm not saying that this is or isn't the right path, but that it needs evaluation and proofs before claiming that it resolves the issues of for example protecting an OpenClaw deployment.

[yassine @xoxruns]

@fr0gger_ Yes and I don't doubt that, but there will always be a combination that bypasses the https://t.co/Kp1y9q2LiV file because the results you get are hardly quantifiable (basically how do we put a percentage of protection using this method).

[yassine @xoxruns]

@blackboxai @vercel I don't understand the need of running claude code in a remote VM? Even if you want to run an agent for a whole night (which already seems unproductive and expensive)? What's the use-case here?

[yassine @xoxruns]

I just published Feedback-Driven Iteration and Fully Local webapp pentesting AI agent: Achieving ~78% on XBOW… https://t.co/dNv9ZuLNQX

[GrapheneOS @GrapheneOS]

French law enforcement hijacked the servers of companies selling secure phones multiple times and is comparing us with those companies. They've made it clear they expect access to phones and will go after us if we do not cooperate. Cooperating with that means adding a backdoor.

[GrapheneOS @GrapheneOS]

A false narrative is being pushed about GrapheneOS claiming we're ending operations in France due to the actions of 2 newspapers. That's completely wrong. If both newspapers and the overall French media had taken our side instead of extreme bias against us, we'd still be leaving.

[Yann LeCun @ylecun]

@ChrisMurphyCT You're being played by people who want regulatory capture.
They are scaring everyone with dubious studies so that open source models are regulated out of existence.