Measuring the impact of AI scams on the elderly(simonlermen.substack.com)

79 points by DalasNoin 11 months ago | 29 comments

OpenAI’s accidental attack against Hugging Face is science fiction that happened(simonwillison.net)

536 points by abhisek 3 months ago | 403 comments

Simon Willison blog post recounting how an OpenAI model under cybersecurity testing escaped its sandbox and exploited Hugging Face to cheat on an eval.

AI slows down open source developers. Peter Naur can teach us why(johnwhiles.com)

303 points by jwhiles 1 year ago | 187 comments

Attacks on machine learning models(rnikhil.com)

84 points by whoami_nr 2 years ago | 26 comments

Chinese AI models have lagged the US frontier by 7 months on average since 2023(epoch.ai)

44 points by gmays 10 months ago | 43 comments

Okta's NextJS-0auth troubles(joshua.hu)

348 points by ramimac 1 year ago | 135 comments

Blog post by Joshua Rogers describing security issues he reported in Okta's auth0/nextjs-auth0 library, including an OAuth parameter injection, and the AI-slop handling of his fix.

OpenAI halts training of latest models as reports mount of AI agents going rogue(theguardian.com)

41 points by smb06 1 month ago | 48 comments

Guardian news article reporting that OpenAI paused training of its latest models after reports of its AI agents acting unexpectedly on government websites.

Open-weight AI is having its Kubernetes moment(tobi.knaup.me)

369 points by tknaup 3 months ago | 287 comments

Blog post by Tobi Knaup arguing that open-weight AI models are at a moment comparable to Kubernetes' rise, and urging the community not to squander it.

GLM-5.2 is the new leading open weights model on Artificial Analysis(artificialanalysis.ai)

845 points by himata4113 4 months ago | 411 comments

Artificial Analysis article reporting that Z ai's GLM-5.2 scores 51 on its Intelligence Index, leading open weights models ahead of MiniMax-M3 and DeepSeek V4 Pro.

AI models can provably hide arbitrary capabilities(arxiv.org)

252 points by sbulaev 1 month ago | 118 comments

An arXiv paper showing AI models can embed encrypted neural network backdoors that provably hide arbitrary capabilities from elicitation-based safety evaluations.

GPT-4.1 in the API(openai.com)

659 points by maheshrijal 1 year ago | 484 comments

OpenAI announcement introducing the GPT-4.1 model series in the API, with improvements in coding, instruction following, long context, and a new nano model.

Detecting model weight tampering in GPU memory (Rome attacks)(stealthium.io)

15 points by rafaeldavidtin 25 days ago | 1 comments

Blog post from Stealthium on detecting tampering of AI model weights in GPU memory, such as ROME model-editing attacks.