Israel's fake thinktank tried to game AI; OpenRouter joins Stripe

Fake US thinktank set up and funded by Israel sought to game AI for propaganda — reports a fake, Israel-funded thinktank built to influence AI systems and online discourse. The mechanism matters more than the politics: coordinated influence ops contaminate web corpora, retrieval sources, evals, and the model-behavior audits you use to check for exactly this.

If your RAG pipeline trusts "authoritative-looking policy institute," that's now a threat model. — n1b0m

OpenRouter is joining Stripe — puts the multi-model router inside a payments company. Plausible logic — routing and billing are the same problem wearing different hats — but if you depend on OpenRouter for provider access, watch what happens to model coverage, billing APIs, and pricing terms. — rvz

Lovable CTO: The Future of SaaS Is Apps That Agents Can Use — has Lovable "branching out from AI-powered web app creation and into MCP-powered 'capabilities'." The thesis: SaaS products ship agent-readable interfaces alongside human UIs, which pushes the hard questions onto API design and permissions rather than pixels. — Richard MacManus

DEEP DIVE

How Claude Watermarks AI-Generated Text — is a 48-minute walkthrough of token sampling, watermark detection, and removal. The removal half is the useful half: it tells you how much weight a provenance claim can actually carry before you build policy on top of it. — Sebastian Raschka, PhD

I built a low-latency AI companion that plays Skyrim with me — is a fun demo and a decent systems case study: latency budgets, multimodal state handling, and a voice loop that has to keep up with a game engine. Real-time agents fail in ways chat agents never do. — pantelisk

Proliferate: open-source, self-hostable Codex for any coding agent — is a self-hostable platform for running coding agents, pitched at teams who'd rather not send their repo to someone else's control plane. The Show HN post is light on implementation detail, so read the code before you plan a migration. — pablo24602

Copyright does not protect AI-generated content in EU — a reminder that ownership doesn't automatically follow from calling a model. If you ship generative features in the EU, work out which outputs are protectable before a customer asks who owns them. — u1hcw9nx

WebMCP: Teaching Your Website to Talk to AI Agents — the same agent-readable-website idea Lovable is chasing, written up as a proposal. Thin on protocol and security specifics, but a fine bookmark if you're mapping the space. — sreenathmenon