The EU already has a paper route to this question that does not run through press statements. Annex XI, Section 1, point 1(d) of the AI Act requires providers of general-purpose models to document the architecture and number of parameters. For models with systemic risk, Section 2, point 3 requires a detailed description of the system architecture explaining how software components build or feed into each other, and point 2(b) of Section 1 requires the key design choices with their rationale. That documentation goes to the AI Office, not to readers, so a depth figure quoted by a chief scientist is not the same artefact.
Thanks for this reference to the Act, that's useful but also limited to information disclosure only. The GPAI Codes of Practice goes further actually. In our most recent report we point to Measure 3.2, which requires: 1) 'at least state-of-the-art' model evaluation and 2) Principle of Innovation, which requires 'equal or superior' safety outcomes from any more efficient architecture. These already imply equivalent reasoning monitorability even if the architecture changes. Our report also goes further into how to operationalize this, but even under the current Code the labs should demonstrate they haven't lost monitorability.
Fair correction on scope. Worth adding that Measure 3.2 sits on a statutory hook rather than floating free: Article 55(1)(a) already obliges providers of systemic-risk models to run model evaluation under standardised protocols reflecting the state of the art, with documented adversarial testing, in force since 2 August 2025. Article 55(2) then makes the Code one route to demonstrate compliance until a harmonised standard is published, and non-signatories must show alternative adequate means to the Commission instead. What is untested is whether reduced monitorability fails that state-of-the-art bar.
It is way cheaper and safer to use programmatic verification and retry with fail than to do any of this. I built and use pathmap(.org) from my 4gb old cell phone for free. I used it and solved spermidine modified ginger EVs a potential delivery system for c9orf72/NEK-1 ALS and I never even took anatomy. (Also solved a potential way to use intradermal hydrogel using exosome-sized ginger EVs to locally deliver therapeutics for breast cancer either to treat or prevent metastasis and injected thermogel with borneal modified ginger EVs + Hyalauranic Acid modified ginger EVs as a likely preferred way to treat endometriosis and ovarian tumors). Since everyone is going bigger, I went decentralized (still building the platform but the core app proves the point).
The EU already has a paper route to this question that does not run through press statements. Annex XI, Section 1, point 1(d) of the AI Act requires providers of general-purpose models to document the architecture and number of parameters. For models with systemic risk, Section 2, point 3 requires a detailed description of the system architecture explaining how software components build or feed into each other, and point 2(b) of Section 1 requires the key design choices with their rationale. That documentation goes to the AI Office, not to readers, so a depth figure quoted by a chief scientist is not the same artefact.
Thanks for this reference to the Act, that's useful but also limited to information disclosure only. The GPAI Codes of Practice goes further actually. In our most recent report we point to Measure 3.2, which requires: 1) 'at least state-of-the-art' model evaluation and 2) Principle of Innovation, which requires 'equal or superior' safety outcomes from any more efficient architecture. These already imply equivalent reasoning monitorability even if the architecture changes. Our report also goes further into how to operationalize this, but even under the current Code the labs should demonstrate they haven't lost monitorability.
https://www.arq.foundation/research/when-ai-stops-thinking-out-loud
Fair correction on scope. Worth adding that Measure 3.2 sits on a statutory hook rather than floating free: Article 55(1)(a) already obliges providers of systemic-risk models to run model evaluation under standardised protocols reflecting the state of the art, with documented adversarial testing, in force since 2 August 2025. Article 55(2) then makes the Code one route to demonstrate compliance until a harmonised standard is published, and non-signatories must show alternative adequate means to the Commission instead. What is untested is whether reduced monitorability fails that state-of-the-art bar.
It is way cheaper and safer to use programmatic verification and retry with fail than to do any of this. I built and use pathmap(.org) from my 4gb old cell phone for free. I used it and solved spermidine modified ginger EVs a potential delivery system for c9orf72/NEK-1 ALS and I never even took anatomy. (Also solved a potential way to use intradermal hydrogel using exosome-sized ginger EVs to locally deliver therapeutics for breast cancer either to treat or prevent metastasis and injected thermogel with borneal modified ginger EVs + Hyalauranic Acid modified ginger EVs as a likely preferred way to treat endometriosis and ovarian tumors). Since everyone is going bigger, I went decentralized (still building the platform but the core app proves the point).