The BaseEssayLessWrong
You can't patch yesterday's AI
In a LessWrong post, Adam Mcmurchie argues that safety is not backdated onto released models, so dangerous capabilities, especially biological ones, will stay available as older models become easy to jailbreak, and reactive legal enforcement will not be enough.
Why it matters
The same thing, for different readers.
For lawmakers
The post argues that reactive legal enforcement is not sufficient for capabilities that stay dangerous once released.
For researchers
A blog argument, with no new data in the excerpt, that safety progress is too slow relative to how long released capabilities persist.
The record
- Questions
- Are AI companies putting stronger safeguards into practice? · Can AI systems act beyond what people intended or allowed?
- People
- Adam Mcmurchie (Wrote it)
- What we read
- We have read only the description the publisher gives. No quotations until someone reads the source itself.
- Found by
- The Radar, run of November 16, 2026, via
lesswrong(tier 1: the source’s own channel or an official record) - Cite this
- LessWrong, “You can't patch yesterday's AI”, November 16, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/lesswrong-2026-11-16-cant-patch-yesterdays-ai
Our notesLight lane: written from the triage excerpt only, nothing fetched, no quotations. The post names current and future models in the excerpt; no claim about them is checked.