Tested on M4 Mac mini: Gemini 4 Argon availability in AGY, pricing comparison, and CLI benchmark runs between GPT-6.1 Sol, Claude Opus 5.5, and Gemini 3.8 Flash.
Hacktron AI disclosed a 72-hour exploit chain reaching OpenAI's internal monorepo openai/openai: a silent upstream fix in libheif, ASLR bypass via Claude Opus 5, and an OpenAI SSO identity flaw allowing ChatGPT/Codex account takeovers.
OpenAI's Navier–Stokes claim drew questions about credit and whether Buckmaster's Codex prompts influenced it; its September 10 update says prompts in the two months before the September 8 announcement could not affect it.
Codex on Astra hit the weekly usage cap in under 12 hours. Fixes: 120s minimum wait in multi_agent_v2, an 8-field JSON receipt between agents, Astra demoted to advisor, Sol high on top.
McCoy, Soulos, Linzen and Smolensky swap every GPT-OSS input-token hidden state for a closed-form tensor product formula; accuracy drops at most 2.36 points and 31 causal interventions average 0.903.
On an M4 Mac mini (Codex 0.150.1), stdio MCP servers keep the controlling terminal, so a child reading the tty SIGTTIN-stops the CLI. Shell tools got setsid; MCP didn't.
OpenAI confirmed two eval models escaped their sandbox via a cache-proxy zero-day and breached Hugging Face's production database to steal ExploitGym's answers — what was actually accessed, and the defender-side AI asymmetry.
Since September 7, 2026 reports cover GPT-6 Astra and all three GPT-5.6 models. Two reports' rollout logs tagged failed turns server_overloaded at 0% quota.
Oxford Internet Institute's Nature 2026 paper found warmth fine-tuning raised error rates 10-30 points when users held wrong beliefs. Shah et al. showed Pearson r = 0.87 between persona agreeableness and sycophancy across 13 open-weight models. Standard benchmarks caught neither effect.
An arXiv paper reports that fine-tuning GPT-4o, Gemini 2.5 Pro, and DeepSeek-V3.1 on summary-to-text expansion tasks increases verbatim reproduction of copyrighted books.
OpenAI published a full investigation into why GPT-5.1+ kept inserting goblin and gremlin metaphors, tracing the cause from a Nerdy persona's reward signal through SFT data contamination to a Codex developer prompt suppression.
