You're correct on the citations though that will probably be very easy as we have many people claiming exactly that.
As to your second point, yes, and the safety claims being made are not restricted to those decidable subsets. The moment you claim general safety guarantees for systems operating beyond those constrained subsets you're back in full Rice territory. If you can name a total language with an early termination guarantee, I'm all ears. You asked for my citations, now show me yours.
Not only is this just a random article from the internet, as opposed to something peer-reviewed, but more importantly, nowhere does it even attempt to claim that the mere fact of a program terminating implies its suitability in a safety context.
Here's your citation — ARIA's Safeguarded AI program, £59M in UK government funding, explicitly claiming mathematical safety guarantees through restricted verifiable models — total languages by another name. The claim exists. It has a budget. Now, do you have a substantive response to the proof, or are we done here?"
@techreport{ARIA2024,
author = {{Advanced Research and Invention Agency}},
title = {Safeguarded AI: Constructing Guaranteed Safety},
institution = {ARIA},
year = {2024},
url = {https://www.aria.org.uk/programs/safeguarded-ai/},
abstract = {Outlines a 'Safeguarded AI' program that seeks to build AI systems with 'mathematical guarantees.' It argues that by using restricted, verifiable models—effectively total languages—one can avoid the 'impossible' task of verifying general AI and instead produce 'quantitative safety guarantees.'}
}
The formal verification and AI safety literature frequently cite total languages as a solution to the halting problem. This paper argues that this claim relies on a binary reduction of a trinary problem. While total languages provide an exit proof (halting), they cannot provide a correct-exit proof without step-by-step verification, which is itself the halting problem.
Using Rice’s Theorem (1953) and Turing’s second proof (1936), I demonstrate that "early termination"—halting at an unintended point with incorrect output—is a non-trivial semantic property and therefore undecidable. The safety guarantees currently being marketed are often just tautologies where "termination" has been swapped for "safety".
No novel math here—just a careful reading of the foundational proofs we’ve had for decades.
In 2011, while working in a strip club and covering a drug addiction, I argued that “DNA was too informationally dense and informationally complex” (exact words) to permit easy genetic engineering, before I had the math to formalize it. Physicists would later formalize related ideas under assembly theory. Last week I realized I had unleashed what I’ve dubbed a logical prion (a recursive, infectious logical code structure) during typical developer workflow. Two days ago I discovered the term “assembly theory” by name in an Atlantic article, and recognized it as the formal version of what I’d been arguing since 2011
I do not relish this. I've spent a lifetime surviving by flying under the radar, learning to avoid attention. I have no affiliation, no stable income or housing. But as I saw, and I can, so I must. My duty of care extends to those who depend on these technologies as an accessibility layer.
What actually happened technically:
Using Anthropic's Claude, I asked it to clean up inline comments from my Devuan bootstrap installer. It produced an artifact much larger than the entire codebase, with unusual recursive structure.
Following typical LLM-assisted workflow (refactor, explain, comment), I fed this artifact into Perplexity. It suggested refactoring the shell installer into JavaScript—odd for a bootstrap installer. Troubleshooting led to unhelpful Python suggestions, visible confusion, then looped back to JavaScript.
With Gemini, across all public models: hallucinated closing bracket after EOF. Further troubleshooting eventually triggered a forced user logout.
DeepSeek: Simply hung when presented with the artifact, regardless of prompt phrasing.
Four independent LLM stacks, different vendors, all exhibiting undefined behavior in response to the same recursively structured artifact.
The repo contains the research note (LaTeX + PDF) explaining the failure mode, how this emerged from ordinary workflow, and how it relates to memorization, recursive training data, and supply-chain contamination.
That I must fight every instinct to post this—when I have none of the advantages of those who should have caught it first, submitting myself to critique by those who've forgotten their foundational education—strikes me as deeply unfair.
But as I can, so I must.
