Simon Willison: AI users should not simply forward LLM outputs without verifying them themselves
In his short post "Don't be a meat proxy", Simon Willison urges AI users not to pass on unprocessed LLM outputs – they should read, verify and rewrite them in their own words as evidence of their own work.
Simon Willison, an independent commentator on the LLM scene, published a short post titled "Don't be a meat proxy" (loosely, "Don't just be a human conduit"), in which he sets out a principle for using AI tools. In his words, it is fine to ask AI (LLM) for anything, but the output must not simply be passed on without further work – the user should read it, understand it, verify its accuracy and then write the response in their own words.
Willison describes this rephrasing in the user's own words as evidence that the user has actually completed the preceding steps (reading, understanding, verification). According to him, this effort is precisely the added value that a person brings to the process – as opposed to merely forwarding text generated by a model mechanically.
The post was shared via the community aggregator Lobste.rs and is listed under the tags definitions, ai, generative-ai, llms and ai-misuse, which suggests that it offers a general definition of desirable behavior when working with AI, rather than a response to a specific incident.
Why it matters
This is practical advice for everyday work with AI tools: mechanically copying and passing on LLM outputs without reading and verifying them leads to the spread of unchecked content and zero added value from the human. For individuals, this means a specific habit – verify and rephrase before passing something on; for companies, it is a prompt to introduce a similar rule where employees work with AI-generated content intended for clients or the public.
Two audiences, two different impacts
What this means
For individuals
A user who simply copies an AI output and passes it on without reading and verifying it adds no value and risks spreading unchecked content; rephrasing it in their own words serves as evidence that they have verified and understood the output.
For a business
If company employees simply forward unprocessed AI outputs (e.g. in client communications or reports), the risk of spreading unverified or incorrect information and reducing credibility increases; the recommendation calls for introducing quality control for AI-generated content.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.