Israel-funded "think tank" publishes AI-written essays to steer chatbot answers
A synthetic think tank bankrolled by Israel has published more than 100 AI-generated articles since launching less than a month ago — and its explicit target is not human readers but the chatbots that answer their questions. First spotted by Politico, the Hanover Institute for Public Policy is run by the American advertising firm Piro, paid for by Israel's government advertising agency through FARA-disclosed channels, and appears designed to feed the models that continually scan the web for answers.
The mechanics are the interesting part. The site publishes about a dozen question-headline articles every few days on Israel, antisemitism and Palestine; an analysis by detection tool Pangram found the tested articles were written entirely by AI, with only the bibliographies human-written. The site ships an llms.txt file — a markdown page whose whole purpose is making a site easier for LLMs to scrape. And the firm behind it openly sells this as a service: Piro's co-founder has described building "AI Story Optimization" on LinkedIn, mapping "every surface the models read" to strengthen "the signals that shape how AI understands" a client's story. Invoices filed under the Foreign Agents Registration Act show $900,000 for the initial build plus $100,000 in monthly fees.
The take: this is what influence operations look like after search stopped being the front door. A state-funded campaign is treating model outputs as a media channel to be bought and optimized — no bylines, AI images, word-salad prose — betting that retrieval systems can't tell consensus from well-placed noise. It is also, notably, fully legal: every article is filed with the Justice Department. As we covered this morning, OpenAI just disrupted a Russian operation running the same playbook covertly; Hanover is simply the registered, above-board version. Detection will not save us here — provenance standards and source-weighting in retrieval are now an information-integrity problem, not just an SEO nuisance one.
What to watch: whether OpenAI and Google treat llms.txt-optimized advocacy farms as spam signals, or leave the door open.
When your assistant cites a think tank you've never heard of, how would you even know it's synthetic? Tell us in the comments.
Keenable, founded by ex-Yandex search leadership, raises $26 million to index the web for AI agents. Accel led the seed, with Conviction Partners participating; founders Andrey Styskin (who previously led search, AI and cloud at Yandex) and German AI scientist Matthias Petri argue chatbots ground better on purpose-built indexes than on consumer search infrastructure. The company claims an index of more than 100 billion documents already used in production by several AI labs and inference providers. Search was built for people skimming blue links; agents need full-document grounding at API speed — that's a real market gap, and VCs are paying seed prices for it.
pgbot puts Postgres monitoring inside AI coding agents — read-only, 5.9 MB, open source. The Go tool surfaced on Hacker News this week and has pulled in roughly 660 stars in two weeks. It inspects database health — locks, cache hits, slow queries, unused indexes — and plugs into any MCP-capable agent as a strictly read-only tool, so an agent can diagnose a production database but cannot write to it. That last design decision is the whole point: agents earn a place in production tooling exactly when they can't break things.
What to watch: whether pgbot's read-only pattern becomes the default contract for every agent-facing ops tool.
Should agents ever hold write credentials against production databases? Tell us in the comments.
Sources: 404 Media · Politico · The Hanover Institute · TechCrunch · pgbot (GitHub)
