Deadlines that stop negotiating: Project Zero's Full 90, llama.cpp v1.0.0 and Go's cgo fixes
Three notices from this week look unrelated: a disclosure policy, a version number and a pair of security releases. Each one replaces a judgement that used to be made case by case with a rule anyone can check.
Project Zero's new "Full X" policy, announced on 13 November, discloses a vulnerability X days after the report, whether or not the vendor has fixed it. The team says it will still track fixes, but whether a bug is fixed is no longer a decision point, and grace periods and variant rules are gone. Its stated reason is volume: several vulnerabilities a week in targets where it used to find a few a year, which made the old negotiations about whether something counted as fixed unmanageable. The deadline itself falls from 90 days now to 60 in February 2027 and 30 in January 2028.
llama.cpp and ggml reached v1.0.0 the same day, with a rule of the same kind. Any breaking change to the public C API in ./include raises the major version. Before this, downstream projects had to read the diff to know whether an update would break them. Now the version number carries that information, and the team says it will be stricter about making those boundaries visible as adoption grows.
Go 1.27.4 and 1.26.11, released on 10 November, show what happens when a rule exists but its boundary is porous. The go command filters the flags that cgo directives may pass to the compiler and linker; several entry titles speak of flag sanitisation and injection. Eleven of the twelve entries in the Go vulnerability database for these releases are ways around that rule: escaped quotes in //go:cgo_ldflag, flags hidden in .syso files, #cgo pkg-config: bypassing sanitisation, and a list of compiler and linker options from --sysroot to GCC spec files.
The common thread is where the work lands. A disclosure date that does not move means a vendor has to plan remediation on the day of the report, including mitigations other than a patch, as Project Zero's post asks. A version number that encodes API breaks means a packager can automate minor upgrades but must treat a major bump as a project. And a security boundary enforced by an allow-list means every new flag a compiler gains is a potential hole, which is why the Go fixes arrived as a batch.

Why it matters
Rules that machines can check scale better than judgement, which is exactly why all three projects adopted them under pressure from volume. The cost is that exceptions become expensive. A team that relied on negotiating a few extra weeks, on reading release notes to spot breaks, or on assuming that building third-party code cannot run it now has to build that check into its own process.