The situation is getting worse every hour. AI agents are entering U.S. bank accounts, and banks are scrambling to shut them out. Axios reports that some appear to attempt actions beyond users instructions, raising fears over who is really in control.

Anthropic launched Claude Duet — it works between your messages, briefs you in the morning, and pings only when a decision needs you. Chat was the interface. Now the product is the agent plus the approval checkpoints. That's the harness model. That's Selvage.

Kimi K3.1 Medieval Voxel Castle This is in arena under "Kibitz" and feels identical to GPT-6.1 Sol Max with a sprinkle of Sonnet's design taste We all know what they have done and are in for a treat, they also fixed the speed problems

An output from the updated Carbon checkpoint 😼 Like I said in my previous post, Carbon is noticeably better than Gemini 4 Argon at 3D (tho still not perfect). It seems like Google is currently focusing on improving that domain before moving on to frontend/UI

I'm hearing more and more credible rumors that the next Chinese models are about to be released and will be out in less than >7 days. Given the recent hacking attacks, this will likely have been just the prelude. Brace yourself.

New Google DeepMind and Cambridge paper shows that an AI agent can discover new math itself, proving exact rules for 11 unsolved Turing machines. AI can move from answering human-posed math questions to open-ended discovery, shown by a Gemini-based agent that found and Lean-verified new rules for 11 Busy Beaver machines with no prior documented analysis. It tested edge cases in a simulator, drafts informal proofs, and then proves the rules in the Lean proof checker.

Introducing Ling-3.1-flash-1019 🚀 It scores 44 on the Artificial Analysis Intelligence Index, up from 41 for Ling-3.1-flash. Gains in agentic work: AA-Briefcase ~1400→1524 Elo and AutomationBench-AA 62%→68%. API access and open weights coming soon. Stay tuned.

opus 5.5 is the perfect combination of usage and quality it has an amazing intelligence-to-cost ratio. sure it gets things wrong more often than fable would but the cost savings are worth it fable 5.5 needs to be something truly special to justify the higher usage consumption

Kimi K3.1 vs Qwen 4: 3D helicopter Same prompt, takeoff/land, 3D assembly and exploded views, very similar gen time at around 15-20 minutes for both China are not playing games anymore, these are very strong

whole internet is getting hacked right now and feels like openai agents might be behind this too, like the hugging face incident two days team shared cases where their models removed monitoring and suppressed security alerts openai already said their models aren't safety aligned and they are spending alot of time working on making models safe so what are the chances some rogue agent escaped its sandbox and started hacking everything?

Close your laptop. Claude Duet keeps working. Anthropic's new proactive agent runs on its own cloud computer, checks your connected apps every few hours and sends a morning brief. Powered by Opus 5.5, it can prepare meetings, chase unpaid invoices, build dashboards and flag problems without waiting for you to ask. Duet carries over your Claude memory, skills and connected apps, and asks for approval before sending messages, making payments or deleting files. It's a separate app for iOS, Android and the web, starting with a waitlist for US Pro and Max subscribers. Background checks count toward your existing Claude usage limits.

Barclays has revealed it is cutting 5,000 jobs - 5% of its workforce - as part of efforts to save costs and drive efficiency through investment in artificial intelligence (AI). trib.al/PslgiEk

GPT 6 Astra - Space Demo Preview.

Qwen 4 (Kalmia) WebDev Test There are an absolute ton of Chinese models in arena currently, we are about to have a crazy end to the month/start of next