Vendor
Anthropic
Anthropic is an AI safety company founded in 2021, best known for the Claude family of AI models. Its Computer Use feature enables AI agents to control computers by screenshotting and acting on what they see — the technology at the centre of this article's privacy and prompt-injection argument.
Products & Services
Computer Use (agentic computer control via screenshots)Branded baseball caps, pop-up merchandiseClaudeFable 5Mythos 5Opus 4.8automatic fallback/billing routing
Coverage in Rack & Reason (13)
- J.Crew's AI Stylist Knows the Shoot Better Than the Shopper
- Muse Lost eBay. Its Robots File Now Lets ChatGPT and Claude Search.
- Seven Thousand Agent Sessions Ended In Seven Thousand Bookings
- Otto Can Replace Any AI Model Except the Router That Picks It
- Claude Got a Seller Central Plugin. Muse Got the Conditions of Use.
- AcCoRD Scored Four Kinds Of Shopper. Agents Failed The One Apparel Sells To.
- The Cache Decides Which Questions a Retail Assistant Can Afford
- Ten Languages Went Into the Benchmark. Commerce Came Out Worst.
- The Lab Has No Store. It Has a $239 Chore Coat.
- Agents Were Handed the Price. The Benchmark Asks If They Can Hold It.
- The Agent Dropped 37% of Its Tools. Success Stayed Comparable.
- A Traitor Agent Passes Every Check You Can Read
- The Shopping Agent Has to See Your Screen. The Retailer Owns What It Sees.
First mentioned: June 2026