Safety lens

Seven questions asked of every entry, about what the instrument says on incidents, uses, models, enforcement, who receives reports, borders and frontier AI. Each coded value names its source.

Lens values as of 27 October 2026.

Safety lens by entry. Unclear means no cited source settles the question yet. Lens values as of 27 October 2026.
Entry	Incident reporting	Unacceptable uses	General-purpose models	Enforcement	Who receives reports	Cross-border	Frontier AI	What would change this row
The AI rules that keep almost arrivingOfficial statement	Unclear	Unclear	Unclear	Unclear	Unclear	Unclear	Unclear	Publication of the draft text.
Three risk tiers, not fourOfficial statement	Yes	Yes	Unclear	Soft	Unclear	Unclear	Unclear	Publication of the draft text.
A 72-hour breach clock, and nobody to callPress report	Yes	Unclear	Unclear	Unclear	Named, not set up	Unclear	Unclear	A full reading of the published text. The named body being set up.
Report everything, not just high riskOfficial statement	Yes	Unclear	Unclear	Unclear	Unclear	Unclear	Unclear	Publication of the draft text.
ASEAN's generative AI guide asks the region to explore incident reportingOfficial statement	Unclear	Unclear	Yes	None	Unclear	Unclear	Yes	A full reading of the published text.
Circular Letter 9/2023 sets AI ethics guidance, with no incident dutyLegal text	No	Yes	No	None	Not named	No	No	A new or amended text.

Definitions

Yes: A cited source states it.
No: A cited source states the opposite, or the full text has been read and it is absent. Absence in a press summary is not no.
Unclear: The default, until a source settles it. Unclear is not no.

Enforcement is binding, soft (for example self-assessment, no sanctions), none, or unclear.

Who receives reports is named and operating (a named body that exists and works), named but not set up, not named, or unclear.

The validator rejects a coded value without a basis sentence, and a "no" taken from a press report.

The draft texts of the AI roadmap and AI ethics regulations were not published as of 26 October 2026: neither appeared on the ministry's official list of draft regulations. Lens values for those drafts come from officials' statements.

How the lens maps to threat models

This mapping uses only the claims of Analysis 01. It shows which questions matter for which kind of risk, not how likely any risk is.

Threat models, first signals and the lens questions that matter most
Threat model, as in Analysis 01	Where the first signal tends to appear	Lens questions that matter most
Misuse at scale	Unusual account activity and intrusions at targets, seen by the developer or the targets.	Incident reporting, Enforcement
Misaligned agent action in the wild	Out-of-scope actions on third-party systems, often seen first by the third party or outside researchers, sometimes abroad.	Who receives reports, Cross-border, Incident reporting
Loss of control	Not observed, and judged beyond current systems. Signals would likely be weak, scattered and partly outside the developer's logs.	Frontier AI, Incident reporting, Who receives reports

Source: Analysis 01, Section 3 and Table 2. The classification is the author's assessment. Read Analysis 01

Safety lens by entry. Unclear means no cited source settles the question yet.

Definitions

How the lens maps to threat models