Google HEIR and Cloudflare's AI security controls

Google HEIR and Cloudflare's AI security controls

Daily AI Tool Drop. On August fourteenth, the most useful launches are about controlling where AI data and actions can go. Google showcased H E I R, a compiler toolchain for encrypted inference. Cloudflare added Access protection for Workers and new controls for M C P traffic. Here's what shipped, who it is for, and the limit on each one.

0:00 / 2:59
Two August 14 launches focus on a practical question for AI teams: how do you keep data private and agent actions inside an intentional boundary?

The briefing

  • Google HEIR: Google showcased HEIR, an open-source compiler toolchain for homomorphic encryption. It is designed to help pretrained AI models operate on encrypted inputs, so a server can compute without seeing the underlying data. Google demonstrated recommendation, fraud-detection, network-anomaly, and hotword workloads. The opportunity is private inference in cloud services; the caveat is real overhead. Google's latency figures are single-threaded CPU results, and accelerator work is still ahead. 1
  • Cloudflare Access for Workers: Cloudflare launched a way to attach an Access policy directly to a Worker. That protection follows the Worker across routes, custom domains, workers.dev, and preview deployments. Policies can be applied at the account or application level, while authenticated identity details are available in code. It targets internal tools that are often deployed before login and authorization are designed. 2
  • Cloudflare MCP traffic controls: Cloudflare Gateway now describes MCP requests using protocol-level heuristics, helping security teams find shadow MCP traffic, allow Portal-only access for approved servers, and block direct connections on managed network paths. This is a governance and network control, not an MCP SDK. It helps organizations inventory and constrain the systems agents can reach. 3

Why it matters

Google is moving privacy into the computation itself. Cloudflare is putting identity and network policy around the applications and tools that agents use. For a safe test, start small and measure three things: what data crosses the boundary, which identity can act, and what latency or review effort the control adds.

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Related content

More from this channel