Mistral got a US patent on tool calls in 118 days. Without prior public notice
berofeev
18 points
9 comments
August 13, 2026
Related Discussions
Found 5 related stories in 51.7ms across 4,128 title embeddings via pgvector HNSW
- Mistral Patent for “Code implemented tool calls” theanonymousone · 137 pts · August 10, 2026 · 81% similar
- Microsoft strikes 'multibillion-dollar' deal with French AI firm Mistral taubek · 41 pts · July 22, 2026 · 45% similar
- Mistral OCR 4.1 spelk · 294 pts · August 13, 2026 · 43% similar
- Samsung in talks to invest in Mistral at €20B valuation petethomas · 40 pts · July 22, 2026 · 43% similar
- Bluesky Trademarks ATProto chaosharmonic · 60 pts · July 16, 2026 · 41% similar
Discussion Highlights (8 comments)
eru
Hmm, article is also AI written. Oh, well.
nicolailolansen
How credible is this report? It seems to be written 100% by AI.
A_D_E_P_T
That article is one of the most obvious examples of "Claude style" I've ever seen. Anyway. Software patents are generally unenforceable. This one is extra unenforceable. It's so weak that they won't even be able to play patent troll and bully small companies with it. Also, I'm sure that somebody has already put in a request with the USPTO for an "Ex Parte Reexamination." This allows for a newly-granted patent to be challenged on prior art or patentability/obviousness grounds. The fee is something like $6000, so it's rare as most patents are hardly worth that much to begin with, but I could see it happening here. In short, this was pointless on Mistral's part and a PR own-goal. Nobody else ought to be concerned.
chrisjj
> So the sandbox does not actually suspend a live interpreter. It is stateless. It fakes a continuation by re-executing the whole block from scratch with previously recorded results substituted in Stateless? How is the recorded results not state?
applfanboysbgon
If you have a point, it isn't helped by an AI-generated article making it out to be the most important, dramatic thing in the history of all of humankind.
tao_oat
For some reason we seem to have more patience for bad writing when it's clearly from an LLM. This article starts > US 12,670,045 B1 was filed on 4 March 2026 and granted on 30 June. The claims are now readable and they are narrow: a stateless sandbox that pauses a code block, ships one tool call to the client, and resumes by replaying from the top. The prior art that bears on that is durable execution, not CodeAct. This is really hard to parse! What's CodeAct? This article doesn't seem worth anyone's time. (it clearly wasn't worth the author's time to write it).
fiatpandas
Junky AI written article. I seriously doubt this was fully proofread by a human.
wirj34iuh324uh2
Here is the patent file wrapper, which contains all docs and correspondence: https://data.uspto.gov/patent-file-wrapper/search/details/19... [free USPTO account required for login]