DeepSeek V4 ships hours after GPT-5.5, and the technical report tells a more interesting story than the benchmark bars. Susan Zhang reads the paper out loud: anticipatory routing, logit clamps, and a training run that kept catching fire at 33 trillion tokens. I walk through what the fragility actually means for anyone planning to finetune on top of it.

On the OpenAI side, GPT-5.5 lands with a quiet thud on Victor Taelin's LamBench. Codex picks up a proper reviewer agent. A plugin called endless-toil makes your editor groan at bad code. Sapiens2 admits it trained on half of Flickr's humans. And Fireship spends a week automating his mom's IT support with a voice-cloned agent called OpenClaw.

— Lenar Kess

End-to-end coverage.

Anthropic's pricing experiment backfires as Claude Code nearly exits the $20 tier. Mozilla patches 271 Mythos-discovered vulnerabilities in Firefox. Sam Altman tweets drunk while Google ships silicon for the agent era.

Claude Code crosses the self-modification threshold while Anthropic can't decide if third parties can use their models. Kimi K2.6 brings Opus-level coding to open source at a fraction of the cost. Tim Cook hands Apple's CEO role to hardware chief John Ternus just as the AI battle intensifies. And the entire agent ecosystem runs on undocumented endpoints that could vanish tomorrow.

Kimi K2.6 just matched GPT-5.4 on SWE-Bench Pro. Open-source models are no longer playing catch-up—they're setting the pace. Meanwhile, Atlassian joins the enterprise data grab, 44% of Deezer's daily uploads are AI-generated, and engineers are warning that agent architectures are repeating MS-DOS security mistakes.

Today’s episode is about where the AI story feels real right now: not in grand claims about instant labor replacement, but in the places where systems meet the world and get weird. We dig into Vercel’s April 2026 security incident, Johann Rehberger’s latest Claude memory-hijack experiment, the ongoing fight over whether LLMs can really reason, the local-model push on Apple Silicon, and the memory supply constraints that may matter more than benchmark drama.

• Vercel’s security bulletin and Guillermo Rauch’s thread: how a compromised third-party AI tool and a Google Workspace OAuth pivot turned into an environment-variable incident, and why the phrase "non-sensitive" is doing a lot of work.
• Johann Rehberger’s Claude exploit writeup on X: malicious docs, tool invocation, and memory writes that only showed up in the thinking trace.
• Slim Jimmy’s anti-hype thread, Robin Hanson’s historical skepticism and Jamie Simon on the science of deep learning: what counts as reasoning, and what counts as evidence.
• Walter Rafelsberger’s local Qwen3.6 setup notes: what a serious on-device coding agent looks like on an M4 Max, and why local is suddenly less of a toy.
• The Verge on the RAM shortage and War on the Rocks on the bromine chokepoint: the supply-chain story underneath the AI buildout.