Patches: measure the enterprise exposure window
The response needs to connect a public advisory to the service users actually reach. A record for each component supports priorities, change testing and ownership of unresolved work.
Czytaj po portugalskuBy Pentesty
Two sources, one operational question
On November 13, Project Zero announced Full 90 as its general rule: disclosure 90 days after reporting, fixed or not, with plans for 60 days on February 2, 2027 and 30 on January 3, 2028. On November 16, GTIG reported exploit_agent configured to analyze patches and execute exploits, with victim data on the observed server.
These are independent publications with different purposes. Our enterprise proposal uses them to review the interval between available information and effective protection. That requires separating a tool's configuration, a recorded attempt and a confirmed effect. A public account of an investigation does not automatically reconstruct what happened to every service in your organization.
Record the three milestones
Choose an important component and create a record identifying its product, owner and maintenance channel. Record the disclosure date stated in the advisory, when the applicable fix became available and when the control was observed in your environment. Preserve the reference and version consulted. Keep any unknown milestone visible instead of filling in an approximate date.
The first date belongs to disclosure coordination; the others also depend on distribution and operations. An available update may still await acceptance testing, synchronization or a restart. Our Linux kernel analysis explains why acceptance needs to identify the component actually loaded. For an application, likewise identify the process, image or instance executing the code.
Prioritize around the service
Connect the component to the functions it provides and the paths through which it can be reached. Examine authentication, effective permissions, the data involved and continuity dependencies. Combine that map with exploitation status and the vendor's current guidance. Identify who can authorize an emergency change, temporary containment or a documented risk exception.
The outcome should be a service-specific decision with an owner and a review date. Avoid a queue that treats every update as interchangeable or closes machines that failed to respond. Separate inventoried assets from those actually checked. When delivery depends on a provider, record which component its commitment covers and which parts still need evidence from your organization.
Validate the change in a controlled setting
In an authorized staging environment, use a representative version, test accounts and synthetic data. Before the change, define a legitimate function that must continue and an operation the control should deny. Collect both outcomes before and after the update. Connect each response to the revision running and the configuration examined, preserving the test's time and identity.
The exercise can verify version, authorization behavior and functionality without installing an offensive framework or reproducing a campaign. A generic request failure does not establish that the intended control worked. If an AI tool proposed the fix, review the change and its assumptions, as in Daybreak Defense Factory acceptance. Record tests that remained outside coverage as well.
Close with evidence for each instance
Expand the rollout in agreed groups with defined stop and recovery conditions. Preserve the observed version, service result and verification time for each instance. Check replicas, persistent nodes and images used for subsequent deployments. An aggregate dashboard helps coordinate the work when it also lets reviewers locate entries without confirmation.
Track the interval between fix availability and verified control, including exceptions that remain open. Explain the measurement's denominator and limitations before presenting an average. At handover, provide the evidence, accepted risk and owner of the next action. The conclusion describes protection observed within that scope and period; investigation of possible earlier access retains its own questions.
Frequently asked questions
Do the 90 days define my internal deadline?
That deadline belongs to the disclosure policy. For your fleet, combine priority, exposure and operating conditions in an explicit decision for each service.
How should an unresponsive asset be tracked?
Keep it in the inventory and unresolved-work denominator. Record an owner, the reason evidence is missing and the next collection attempt; no response does not close the item.
Does an automation configuration prove compromise?
Require separate evidence of configuration, attempt, access and effect. A decision about your environment needs your own records and an explanation of visibility limits.
Can an AI-proposed fix be accepted directly?
Define review, authorized tests and expected functionality before acceptance. Preserve the code evaluated and the results, alongside the tool's summary.
Sources and references
- Project Zero: policy announcement, November 13, 2026
- Project Zero: current policy and exceptions
- Project Zero: disclosure questions
- GTIG: agentic threat activity report, November 16
- Help Net Security: Project Zero policy coverage
- Infosecurity Magazine: disclosure schedule
- The Register: exploit_agent coverage
- SiliconANGLE: GTIG report coverage
- BleepingComputer: GTIG report coverage
Original Pentesty analysis, published November 16, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.