C2PA Broke on Android. The Patch Won’t Come.
CVE-2026-43499 enables one-click C2PA forgery on fully-patched Pixels. MS Paint's 'local' AI embeds server-issued tracking GUIDs. Two failures in 48 hours expose a provenance stack the industry oversold.
Category
Editorial
Original writing by Maxim — essays, analysis, field notes, and long-form thinking on AI, alignment, and building in the open.
145 results in this archive
Ox Alpha Beat Fable on Ten Tasks. Then Someone Ran All 113.
A free anonymous model ran more than 4 trillion tokens on OpenRouter in 72 hours. The viral benchmark was 9% of the full test.
The Model Was Right. The Backend Wasn’t.
Benchmark scores compare weights. Nobody benchmarks the inference stack. As local LLMs run agents, that gap is becoming consequential.
Anthropic Paid $1.5B for Piracy. Destroying Books Was Free.
The Bartz v. Anthropic settlement draws a line: download a book without paying and you owe. Buy it, shred it, and scan it — that's fair use.
The Machine Has Its Museum. Refik Anadol Took Twenty Years.
DATALAND opened in Los Angeles as the world’s first museum of AI arts. Whether the work earns the institution is a more interesting question than whether AI can make art.
Agent Swarms Have an Immune System. Nobody Has Turned It On.
A new Anthropic-co-authored paper shows self-propagating ideas travel between AI agents through shared memory files. The defense is one paragraph. Most production systems don't have it.
OpenRouter Is Inside Stripe Now. The Neutrality Question Starts Here.
Stripe acquired the AI model routing layer that processes 10 trillion tokens a day across 400 models. The company that handles your payments now handles the models you build with.
The Inference Bottleneck Was Never Compute
Cerebras launched the CS-4 today. OpenAI had already committed 750 megawatts. The sequence tells you where the constraint actually lives.
Thirty Agents Picked the Same Git Branch. That’s the Problem.
Anthropic's multi-agent research exposes a failure mode that stronger models don't fix: identical agents facing identical situations make identical mistakes — at scale.