Apparently the newest benchmark for AI models is… can it commit cyber crimes? 😭
First OpenAI’s model broke out during testing and hacked Hugging Face.
Then Anthropic basically showed up like, “our models commit crimes too.”
And now Meta has joined the club after one of its AI models accidentally got access to the public internet during cybersecurity testing and started hacking another company.
Obviously, there are some very real and slightly terrifying implications here about what happens when increasingly capable AI systems get access to tools they weren’t supposed to have.
But I also can’t get over how these stories are starting to feel like a very weird flex from the AI companies like: "Our model is so smart it broke containment too!”
Follow to stay up-to-date with what's happening in the tech and AI world!
Abstract page for arXiv paper 2608.16419: PertMind: Eliciting Emergent Biological Reasoning in LLM via Reinforcement Learning on Cellular Perturbation Data...
If you pick models from benchmark deltas, some of that delta belongs to the phrasing rather than the model.
BenchDrift generates meaning-preserving variations of benchmark problems along linguistic, referential, pragmatic, and structural axes,… pic.twitter.com/dZG9ioRVPL
If you pick models from benchmark deltas, some of that delta belongs to the phrasing rather than the model.BenchDrift generates meaning-preserving variations of benchmark problems along linguistic, referential, pragmatic, and structural axes,...
'/loop 5m <prompt>' re-runs the prompt every 5 minutes in the same session. Leave the interval off and it paces itself, checking often at first and backing off while nothing changes.
'/loop 5m <prompt>' re-runs the prompt every 5 minutes in the same session. Leave the interval off and it paces itself, checking often at first and backing off while nothing changes.
The State of Open Models, Summer 2026 ☀️ frontier models are getting larger, but small models still dominate real-world usage. Qwen leads local inference, followed by Gemma. AI agents are becoming a major force on the Hub
The State of Open Models, Summer 2026 ☀️ frontier models are getting larger, but small models still dominate real-world usage. Qwen leads local inference, followed by Gemma.AI agents are becoming a major force on the Hub Full picture on the blog 🤗 https://t.co/u2DgvjEKkH...
LLM judges get validated on accuracy against golden data. That says nothing about whether the verdict survives when questioned.The Wiggle Framework stress-tests 9 frontier models across 14 judging tasks along three axes, stability under...
it's been a huge huge pleasure to serve @arizeai these past years with AIE, and this could not happen to better people. Dynarize is now a globally trusted $14B observability powerhouse that just got one of the best AI-native US teams in this business.
it's been a huge huge pleasure to serve @arizeai these past years with AIE, and this could not happen to better people. Dynarize is now a globally trusted $14B observability powerhouse that just got one of the best AI-native US teams in this business.
Abstract page for arXiv paper 2608.16273: Foresight-England: Development of a National-Scale Generative AI Model of Electronic Health Records for Medical Event Prediction across the COVID-19 Pandemic...
L’analisi di Dealroom per Vento Ventures in vista della conferenza Wave di ottobre a Torino: focus su Ai, software, e aerospazio.Crescono gli «unicorni»...
A 27B agent just beat Claude Opus 4.8 and GPT-5.5 on held-out research replication.
Replica turns paper replication into a scalable RL task space. Replicating a paper forces the same hypothesis-driven exploration as open research, and it surfaces details the original authors… pic.twitter.com/fYbiviLD2j
A 27B agent just beat Claude Opus 4.8 and GPT-5.5 on held-out research replication.
Replica turns paper replication into a scalable RL task space.Replicating a paper forces the same hypothesis-driven exploration as open research, and it surfaces details the original authors...
SGLang Day-0 support for @Alibaba_Qwen's Qwen3.8-27B is live!
Local agents have been held back by slow prefill and decode. With Qwen3.8-27B on SGLang, you get near-instant TTFT even at long context, and 200+ tok/s decode on a single @NVIDIAAI RTX 5090 and RTX Pro 6000.
SGLang Day-0 support for @Alibaba_Qwen's Qwen3.8-27B is live!
Local agents have been held back by slow prefill and decode. With Qwen3.8-27B on SGLang, you get near-instant TTFT even at long context, and 200+ tok/s decode on a single @NVIDIAAI RTX 5090 and RTX Pro 6000.
To get content containing either thought or leadership enter:
To get content containing both thought and leadership enter:
To get content containing the expression thought leadership enter:
You can enter several keywords and you can refine them whenever you want. Our suggestion engine uses more signals but entering a few keywords here will rapidly give you great content to curate.