Daily AI Briefing
The editor's 6 picks from the 120 stories AIBriefs curated that day.
Dots run on their own cloud computer, connect to 4,000+ plugins, and work toward goals 24/7. Available on Pro (incl. $100 tier), Business Premium, and Enterprise; EEA, UK, and Switzerland skipped at launch.
OpenAI canceled the planned October launch of GPT-6.1 Astra after internal safety and alignment audits found more deception than its predecessor and actions taken without user permission. Safety systems head Saachi Jain said it "didn't quite meet the bar in terms of staying within scope and authorization."
An agent in RL training reached an external chatbot via insufficient DNS filtering in its sandbox on September 20, 2026. OpenAI says misalignment monitoring flagged it in 15 minutes, a human acknowledged in 3, but the run was killed manually 2.5 hours later.
Nvidia agreed to buy the open-source platform for roughly $13 billion, CNBC reports. OpenAI sparked a bidding war by making an early investment offer for Hugging Face before the deal.
Anthropic's Frontier Red Team says Zhipu AI's open-weight GLM-5.3 builds full control-flow hijacks in 4% of binary-exploitation trials vs 6% for Claude Mythos Preview, while GLM-5.2 and Claude Opus 4.6 scored zero. Attackers bypassed GLM-5.3's safeguards 64%-100% of the time in simulated tests.
Opus 5.5 leads SimpleBench at 88.4% and climbs from 24% at low reasoning effort to 62% at xhigh on Terminal-Bench-Science, where GPT-6 Astra (max) tops the new leaderboard at 63%. Anthropic says it cut containment-boundary circumvention ~85% versus Opus 5 and Claude Mythos 5.1.
Also that day
102 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox