FrasiertheLion

[1] https://tinfoil.sh/blog/2026-10-25-less-is-more

Of course, we can't support Claude or Grok as they are closed source, but there is no incentive for companies that need your data to train the next generation of models to allow for private inference. One day...

>that open models are in the ballpark of the best commercial models

This is basically true for certain tasks. As an example, chat interfaces are not well poised to take advantage of higher model intelligence than what the best open source models already provide. But coding harnesses still benefit from greater model intelligence and even more so, the reinforcement learning that tightly interlinks the provider's coding harness (claude-code, codex) with the model's tool calling interfaces is another reason for discrepancy in effectiveness even when controlled for model intelligence. The opencode founder (open source coding harness that supports different model providers) was recently complaining about the challenges making the harness work well with different providers: https://x.com/thdxr/status/2053290393727324313

A useful framing over “local vs cloud AI” can be split along two axes: does the task touch private data, and does it need frontier intelligence? You can use frontier models for developing the software (doesn’t touch data), but open-source models running locally for ops: maintenance, debugging and monitoring (touches data). If you need to fall back to frontier intelligence at some point for a particularly hard to resolve problem, you can still rely on local models for pre-transforming and filtering input in a way that's privacy-preserving or satisfies some constraint before it’s sent off to the cloud for processing. OpenAI's privacy filter is a good example of a model that can be used to mask PII and secrets and that can run locally: https://openai.com/index/introducing-openai-privacy-filter/, before sending any data externally for processing.

Another framing for local vs frontier closed which the article mentions is whether the task saturates model capability. With certain tasks like PDF processing or voice or summarization, adding more intelligence isn't necessarily useful. Arguably we've approached that point for chat interfaces already with frontier open-source models. But for coding and ops through well structured tool use inside a coding capable harness, we're still a ways away.

Tangentially, a contrarian take here is that AI can actually enable more privacy preserving software if you’re so inclined. You can just build personalized software and it lowers the barrier to entry and the effort required to self host. SaaS complexity often comes from scaling and supporting features for all types of customers, and if you're building software for personal use, you don't need all that additional complexity. Additionally, foundational and infra software that is harder to vibecode with AI is often already open source.

The technical walkthrough section on this blog that we co-wrote with one of our customers walks through the various attack surfaces: https://www.workshoplabs.ai/blog/private-post-training

We weave in many mitigations against attacks, but it depends on what class of attack it is.

If there are specific attacks you are concerned about, happy to provide an answer if it’s something we can address or not.

In turn, that attests the model enclaves, for instance, see https://github.com/tinfoilsh/confidential-deepseek-v4-pro. The model repo/release that the model router attests is included in the attestation config, which creates a chain of trust.

Also see https://docs.tinfoil.sh/verification/attestation-architectur...

I wrote a blog about this: https://tanyaverma.sh/2026/03/01/nowhere-to-hide.html