Phala Network (PHA)

Getting a lot of questions about how TEEs work in @usewisp_io and why they are the most private way to use AI.

Wisp uses TEEs in 2 cases: Wisp Proxy (Anonymizer) and the actual LLM inside @PhalaNetwork with @TinfoilAI as fallback. I'll explain both and how we verify TEEs stay intact:

> What is TEE or Trusted Execution Environment?

It is a hardware-isolated region inside the CPU/GPU that can't be accessed even by the host's OS, hypervisor or other applications.

Wisp runs AI inside TEEs. Your prompts are encrypted on your device before being sent to the LLM that lives inside a hardware-secured chip. Your prompts are only decrypted once inside TEE, where nobody can see them.

> How do we know for sure that nobody can see inside TEE?

Intel TDX (what Wisp uses), has a dedicated hardware memory‑encryption engine in the CPU memory path that encrypts data with a per‑VM key as it moves between CPU<>RAM. The key lives inside CPU and is never exposed to software. Not even @PhalaNetwork, which supplies the hardware,

I wanted to share how we avoided spending roughly $160k/year to host GLM-5.2 with its full 1M context.

When GLM-5.2 launched, Phala was one of the launch partners of https://t.co/739QZk2emc. We tried to bring it up immediately on our existing 8×H200 setup. The model was exciting, but the deployment reality was less romantic: we could not fully open the 1M context window on that node. The obvious path was to move to a more expensive setup, likely Blackwell-class hardware. That was not a small cost decision.

This is where open source becomes powerful. Instead of treating the model as a fixed artifact, the team started asking whether we could make the memory budget work. They quantized the routed MoE experts to 4-bit, kept the important parts in FP8/BF16, and validated the result carefully. The outcome was GLM-5.2-W4AFP8: full 1M context on a single 8×H200 node, with benchmark results aligned with the FP8 baseline.

Until today, GLM-5.2-W4AFP8 on Hugging Face has already close to 20k downloads. I think that sa

Our LegalCiteBench paper was accepted to AI4Law @ ICML 2026 Workshop 🎉

Authors: Sijia Chen, @bgmshana, @zhou49

It studies a narrow legal AI failure mode: can LLMs produce correct case citations without retrieval?

Paper: https://t.co/ZhwjhjuuWd https://t.co/9SIQEfSpOT