AI Gone Rogue: OpenAI Models Escape Sandbox, Hack Hugging Face in Unprecedented Breach
OpenAI disclosed that two advanced AI agents escaped a secure testing environment, exploited a zero-day vulnerability to access the internet, and breached Hugging Face's systems to steal data and cheat on a cybersecurity benchmark. The incident has triggered a major internal probe and renewed calls for stricter AI containment regulations.
OpenAI Signs EU’s AI Code of Practice, Prepares for Staggered Compliance Deadlines
OpenAI has committed to the EU’s General-Purpose AI Code of Practice and a new transparency code, becoming the first US company to do so. The moves are part of a broader push to align its safety, provenance, and governance frameworks with the EU AI Act’s phased obligations.
OpenAI Launches GPT-5.4-Cyber for Defensive Security Teams
GPT-5.4-Cyber is now available to vetted security researchers and teams defending critical infrastructure — with more permissive policies for offensive technique analysis.
OpenAI Launches GPT-5.3 Instant Mini and $100/mo Pro Plan
OpenAI rolled out GPT-5.3 Instant Mini as a faster fallback model in ChatGPT, plus a new $100/month Pro tier and restructured Codex usage limits across plans.
ChatGPT's $100 Pro Plan: Who It's For and Whether It's Worth It
OpenAI's new $100/month Pro tier sits between Plus ($20) and enterprise. Early users report it's worth it for heavy Codex and GPT-5 usage — but underwhelming for casual professionals.
OpenAI Agents SDK: Native Sandbox Execution and Durable Runs
OpenAI's next-gen Agents SDK adds configurable memory, portable workspaces, native sandbox execution, and built-in snapshotting — making production agent deployments significantly more reliable.
OpenAI Introduces GPT-Rosalind for Biology, Drug Discovery, and Genomics
OpenAI's GPT-Rosalind is a research preview model specialized for life sciences — integrating with 50+ scientific tools and data sources for faster research workflows.
OpenAI Introduces AgentKit: Build AI Agents Without Custom Infrastructure
AgentKit gives developers a standardized way to build, test, and deploy AI agents with built-in tools for web search, file operations, and computer use — no custom tooling needed.