The Ministry of Science and Technology’s criteria for public-service AI platforms, in force since 19 June, mark a platform out of 100 and come in two blocks. Thirty-five marks are for compliance: obeying the law, citing sources, staying inside the data it was given, security, personal data, ethics. Sixty-five are sector-specific, labelled for the work itself: synthesis, analysis, legal interpretation, operational guidance, with separate criteria for fourteen ministries, three agencies and thirty-four localities. Passing takes 70 overall and, separately, at least half the compliance block.
On 14 August the technology press reported an enforcement directive barring classified documents from public platforms, and named two domestic systems currently authorised for state legal work. The sheet itself came up here on 17 August, without its weighting.
What those sixty-five marks reward, quality of output or presence of a feature, is not published, so the split between the two blocks is all the ministry has disclosed. The point values inside each block are unpublished too, so a total does not tell an outsider which block it came from.
Ten pilot projects across four cities take submissions until 30 August and sign before the end of September. Which platforms clear the bar is the first thing anyone outside the ministry will see. Whether the sector block or the compliance block decided each of them is not something the score will show.
Assorted links
-
Tussell counts 2,129 artificial intelligence contracts awarded across the UK public sector since January 2018, worth around £5 billion, and says openly that deciding what counts as AI is subjective. The figure counts contracts, not money spent.
-
India’s Independence Day speech on 15 August was translated live into 22 languages by Bhashini, an open-source translation stack the Indian government owns and runs.
-
A new paper measures a government’s digital capacity by counting the public internet addresses its agencies hold, in a panel of more than 150 countries from 2019 to 2024, and 1,681 subnational regions observed in every year.