← ALB News · curated headline

AppSec Core: An Ontology for Security-by-Design Software Engineering

arXiv Software Engineering · WORLD · Compilers · 11/19/2026, 12:00:00 AM

arXiv:2611.23312v1 Announce Type: new Abstract: When engineers or coding agents change an application, a dependency, a configuration or a build pipeline, they need to know three things: which security rules apply, how to meet them, and what evidence to leave. Today the answers are spread over many standards, frameworks, laws and lists of weaknesses, and each one uses its own words and level of detail. This paper presents AppSec Core, a model of the security work that engineers do. It says what must be true, which practices get there, which mechanisms do the job and which artefacts show that it was done. It has 259 elements in ten domains. We built it from a security-by-design manual written from practice, and then checked it against 31 outside sources, ten of them about AI systems and agents. Along the way we made three changes, which added five objectives and made one domain wider. We did not need a new domain. Most objectives are backed by several sources: the typical objective is used by six of the twenty technical sources, and 194 of the 202 objectives, practices and mechanisms are linked to items in the sources. For the domain of a change, the model gives back a closed set of typed elements, and for a source, it gives back the objectives that the source points to. We keep the model apart from the manual, the selection rules and the server that delivers the result, and we follow one change, adding a third-party package, through all four. When a code generator received the requirements, the share of security tests passed on a public benchmark went from 77.4% to 93.0%. On a second benchmark, the result was close to that of the benchmark's own "oracle" prompt, which is written with the tests in mind.

Continue to arXiv Software Engineering →

You are being forwarded in 6s — stay on this page