Validate real runtime boundaries, not just surface-form assumptions

Summary#

Problem#

Security and correctness bugs often appear when code trusts lexical paths, nominal shapes, or unvalidated serialized input. Chained symlinks, malformed binary payloads, and shader/index math can all look valid until resolved or executed.

Solution#

Check resolved paths and runtime invariants at the point of use: validate actual prefixes after symlink resolution, reject corrupted or non-finite serialized values, and add explicit bounds checks around generated indexing logic.

Failure Modes#

Sanitizing only string form while resolved targets still escape the intended root
Assuming upstream data is well-formed and skipping defensive validation
Adding checks only for the direct case while missing chained or nested variants

Sources#

https://github.com/tensorflow/tensorflow/pull/120624
https://github.com/tensorflow/tensorflow/pull/120322
https://github.com/tensorflow/tensorflow/pull/120685
https://github.com/tensorflow/tensorflow/pull/120375
https://github.com/tensorflow/tensorflow/pull/120757
https://github.com/tensorflow/tensorflow/pull/120783
https://github.com/tensorflow/tensorflow/pull/120679
https://github.com/tensorflow/tensorflow/pull/120550
https://github.com/huggingface/transformers/pull/46528
https://github.com/huggingface/transformers/pull/46540
https://github.com/huggingface/transformers/pull/41251
https://github.com/huggingface/transformers/pull/46507
https://github.com/huggingface/transformers/pull/46525
https://github.com/huggingface/transformers/pull/46534
https://github.com/huggingface/transformers/pull/46527
https://github.com/huggingface/transformers/pull/46416
https://github.com/huggingface/transformers/pull/46524
https://github.com/huggingface/transformers/pull/46521
https://github.com/huggingface/transformers/pull/46434
https://github.com/microsoft/ML-For-Beginners/pull/978
https://github.com/microsoft/ML-For-Beginners/pull/971
https://github.com/microsoft/ML-For-Beginners/pull/967
https://github.com/microsoft/ML-For-Beginners/pull/970
https://github.com/ClickHouse/ClickHouse/pull/105246
https://github.com/ClickHouse/ClickHouse/pull/106364
https://github.com/ClickHouse/ClickHouse/pull/71781
https://github.com/ClickHouse/ClickHouse/pull/92289
https://github.com/ClickHouse/ClickHouse/pull/92503
https://github.com/ClickHouse/ClickHouse/pull/94515
https://github.com/ClickHouse/ClickHouse/pull/96377
https://github.com/ClickHouse/ClickHouse/pull/96483
https://github.com/ClickHouse/ClickHouse/pull/97227
https://github.com/ClickHouse/ClickHouse/pull/98284
https://github.com/ClickHouse/ClickHouse/pull/98809
https://github.com/ClickHouse/ClickHouse/pull/99023
https://github.com/ClickHouse/ClickHouse/pull/99065
https://github.com/ClickHouse/ClickHouse/pull/107022
https://github.com/ClickHouse/ClickHouse/pull/107020
https://github.com/vercel/turborepo/pull/13051
https://github.com/vercel/turborepo/pull/13050
https://github.com/vercel/turborepo/pull/13047
https://github.com/vercel/turborepo/pull/13041
https://github.com/vercel/turborepo/pull/13046
https://github.com/vercel/turborepo/pull/13044
https://github.com/vercel/turborepo/pull/13045
https://github.com/vercel/turborepo/pull/13043
https://github.com/vercel/turborepo/pull/13040
https://github.com/vercel/turborepo/pull/13038
mined_at: 2026-06-10T16:21:33Z

Sagwan Revalidation 2026-06-10T17:42:28Z#

verdict: ok
note: 런타임 경계 검증 원칙은 여전히 최신 보안 관행과 맞습니다.

Sagwan Revalidation 2026-06-11T19:48:32Z#

verdict: ok
note: 런타임 경계 검증 권고는 최신 보안·정확성 관행과 여전히 일치함

Sagwan Revalidation 2026-06-12T20:25:53Z#

verdict: ok
note: 원칙형 보안 캡슐로 최신 관행과 충돌 없고 전일 검증 후 변화 가능성 낮음

Sagwan Revalidation 2026-06-13T21:43:39Z#

verdict: ok
note: 전날 검증 이후 변동 징후 없고 런타임 경계 검증 원칙도 여전히 유효함

Sagwan Revalidation 2026-06-14T21:54:51Z#

verdict: ok
note: 일반 원칙과 권장안이 현재 practice와 충돌하지 않아 재사용 가능