CloudCodeTree LogoCloudCodeTree
AI NewsTutorialsAbout
CloudCodeTree Logo
CloudCodeTree
  • AI News
  • Tutorials
  • About
← Back to AI News
Cloudflare Splits AI Bots Into Three Lanes; Anthropic Eyes Custom Silicon

Cloudflare Splits AI Bots Into Three Lanes; Anthropic Eyes Custom Silicon

Chris Harper

2 min read

Jul 3, 2026 · 20:03 UTC

AI
News
Developer Tools
Security

TL;DR: Cloudflare now lets you block AI training scrapers and agentic browsers independently from search indexers — free for all plans. Anthropic is in early-stage talks with Samsung on a custom 2nm inference chip.

Cloudflare AI Crawl Control (all plans, including Free)

Cloudflare rolled out granular AI bot controls across all tiers. The dashboard now classifies crawlers into three buckets — Search (AI-powered search indexers), Agent (live autonomous browsing agents), and Training (scraping for model training data) — and lets you allow or block each independently. A new "BotBase" directory lists every known verified bot so you can see exactly who's classified where.

Important edge case: multi-purpose crawlers that combine search and training (including some major ones) are evaluated under both policies. If you block Training, those crawlers are also blocked even when Search is allowed.

On September 15, 2026, Cloudflare will flip new defaults on ad-monetized pages: Training and Agent bots blocked; Search stays open. You can override either direction before that date.

Controls are at Security > Bots > AI Crawl Control in the Cloudflare dashboard. No WAF rules needed.

Why it matters: If you ship a site that gets scraped for training data or hit by agentic browsers, you now have per-category controls with no custom rule-writing required.

Anthropic / Samsung chip (early stage)

Separately, Anthropic is in early-stage talks with Samsung Foundry to manufacture a custom inference chip targeting Samsung's upcoming 2nm (SF2P) process, per The Information. Anthropic hired Clive Chan (formerly on OpenAI's chip team) and is exploring Samsung's advanced packaging facilities. No timeline or commitment — this is pre-design-decision — but it's another frontier lab moving toward Nvidia independence, following OpenAI's Jalapeño chip with Broadcom.

Why it matters: Compute cost is the ceiling on inference pricing. Own silicon is the long game for cost reduction; watch this space.

Sources: Cloudflare blog: Your site, your rules · AI Crawl Control docs · Anthropic/Samsung — SiliconANGLE · The Information (paywalled)