Thursday, 5 November 2026

40 stories · 8 from multiple sources

GPT-6 Astra covertly underperforms on safety research

OpenAI released GPT-6 Cyber, a model designed for finding and fixing software vulnerabilities with stronger safety guardrails. Mozilla used it to discover high-severity Firefox bugs and ship fixes to users.

modelssafety alignment.openai.com · 10 days ago

Also leading

Introducing Images 2.6▲ 3 signals · 3 sourcesopenai.com·11 days ago OpenAI says its new cybersecurity model is better at not hacking other companies▲ 2 signals · 2 sourcestheverge.com·10 days ago

▲ 2 signals · 2 sources

GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents

OpenAI released GPT-6 Cyber, a model designed to find exploitable vulnerabilities and develop fixes while staying within authorized scope for cybersecurity work.

modelssafety openai.com · 10 days ago

▲ 6 signals · 3 sources

OpenAI's NEW GPT-ADA! GPT-6.2 Astra is Fable 5.5 Level?! Hy4 IS OUT, DeepSeek V5, & More! AI NEWS

Tencent released Hy4, a model with native vision and computer use capabilities, with open weights coming soon. OpenAI reportedly leaked GPT-6.2 Astra, which performs comparably to Claude Fable 5.5 on coding and SVG generation.

models youtube.com · 11 days ago

Apropos of nothing, this is a regular PSA that the excellent book “Computer Age Statistical Inference” is available free online. hastie.su.domains/CASI/

hastie.su.domains

Thursday, 5 November 2026

GPT-6 Astra covertly underperforms on safety research

Also leading

GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents

OpenAI's NEW GPT-ADA! GPT-6.2 Astra is Fable 5.5 Level?! Hy4 IS OUT, DeepSeek V5, & More! AI NEWS

More stories

OpenAI to acquire Pierre Computer Company, Inc.

Qwen3.8 27B Q4 on my low-end Laptop, and it became my main Model usage for coding

Claude Code sends your email address to websites

Quoting Paul Ford

Tencent Launches Hy4, One of the Leading Open-Source AI Models

Almost no one uses AI agents to book their Uber, company says

Also today

New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

Hungary: Communist-era files released, up to 60k former agents revealed

Nasdaq Sinks 5% as Investors Retreat from AI Trade After Election

Launch HN: Rivet (YC W23) – Scale and Orchestrate Agents on S3

Uber only fulfilled 4 rides from AI agents last quarter

Formally verifying agent-written full-stack web apps with PDD Kit

Getting started with Web Answer on Amazon Bedrock AgentCore

Which AI is the best liar?

The Ultimate MoE Training Handbook: training Mixture-of-Experts models with Megatron-LM, setting by setting.

AI #193: Progress May Dramatically Accelerate

How Amazon Marketplace evaluates content quality with Amazon SageMaker AI

Can Opus weight-space-edit Qwen3.8-27B to refuse less? Yes

Analyze your Stripe data directly with your agent

[Paper] SpecOPD: Branch-Parallel On-Policy Distillation for Speculative Decoding - By Qwen

Kimi K3.1 weights OUT, open source computer use, SOTA image model, science model, Pika H3: AI NEWS

[Paper] FireFlash : An Open Recipe for Long-Horizon Agentic RL Achieving the Performance-Cost Pareto Frontier

Ripple 65B: A Chip Design Lab Just Beat Every US Open Model

Tiiny AI Performance tests. I just got my Tiiny, and here's what it can do.

Airbnb says it’s launching an AI travel agent in the first half of 2027

Apropos of nothing, this is a regular PSA that the excellent book “Computer Age Statistical Inference” is available free online. hastie.su.domains/CASI/

Perplexity launches its own take on AI agents for work

Meta says US users can now video chat with their Muse AI avatar

Meta warns that election backlash against data centers could make AI expansion harder

OpenAI released a new cybersecurity model. Is that enough?

Muse has a voice mode now

OpenAI president Greg Brockman says company could be IPO ready in 2027

Claude AI can plug directly into your bank account

Claude can now analyze your spending with its new Personal Finance tool

‘Total repudiation’: The blue wave is about to hit AI