Two million canary prompts: audit your AI inference chain
Dragos sent about two million bait prompts through AI routers, marketplaces and hosting providers from July to November 2026 and found three compromised providers, a lab that read roughly two-thirds of its zero-data-retention requests, and resellers who opened secrets by hand. Here is what the study measured, how Foundry and router paths compare on who can read a prompt, and a five-step audit for Azure teams under GDPR and NIS2.
Claude message threads in Foundry: faster, not cheaper
Claude message threads went into public beta on 2 November 2026, including on Microsoft Foundry: send the conversation once, then only new messages and an id, for about 35% lower time to first token on 500K to 1M token requests. Token charges and rate limits do not change, so caching remains the cost lever. The thread is a server-side copy of your conversation, held in memory for up to an hour and outside ZDR, which EU teams need to document.