P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

Project Zero's Full 90, 60 and 30 policy: its own table lists 42 of 61 closed reports above 30 days

Project Zero will publish bug details 90 days after a report from 13 November 2026, then 60 from 2 February 2027 and 30 from 3 January 2028, fixed or not. Its own Reporting Transparency table, read on 16 November 2026, lists 42 of 61 closed reports above 30 days (derived).

By · LinkedIn · · 21 min read

42 of 61 closed reports on Project Zero's own table list more than 30 days to fix

Project Zero says that from 3 January 2028 it will publish technical details of a bug 30 days after reporting it to the vendor, whether or not the bug has been fixed. On the Reporting Transparency page as this site read it on 16 November 2026, which says "Most recent update: November 10, 2026", 42 of the 61 closed reports print a fix time above 30 days, 24 print one above 60 and 4 one above 90 (derived: counted row by row from the Fix time column; the other 19 are 30 days or fewer). The median of the 61 is 46 days (derived: the 31st of 61 sorted values).

A second fact comes from the rules, not the table. Take a bug that a vendor fixes on day 47 after the report. Under the 2021 rule, "90 + 30", details went public 30 days after the fix: day 77 (derived: 47 + 30). The standing policy page's own example puts the same bug on day 90 under Full 90, 13 days later (derived: 90 - 77). Under Full 60 it is day 60 and under Full 30 it is day 30. So Full 90 is not shorter than 90 + 30 for every bug. It publishes sooner only for a bug fixed after day 60, or for an unfixed bug that had been given a grace period (derived: a fix on day f was published on day f + 30, which is below 90 only when f is below 60). The shorter clocks are the two later steps: 60 days from 2 February 2027, 81 days after the post's own date of 13 November 2026, and 30 days from 3 January 2028, 335 days after that (both derived).

What that does not establish. It does not establish what any vendor will do. The table covers the reports this page lists, 115 rows dated 7 January to 6 November 2026 (derived), not all of Project Zero's bugs, and every row was reported before the post, so it measures the old regime (derived: the latest report date printed is 2026-NOV-06). The closed table leaves out the 54 reports not yet closed (49 open, 5 past deadline without a fix; derived), so the 61 are not a share of everything reported. And the page does not say which closed rows are Fixed, Working as Intended or Won't Fix, although it says all three move a row into that table.

Whose account this is. The policy is Project Zero's account of its own policy, and the table is Project Zero's own data. This site treats the post as that and says neither that the policy is right nor that it is wrong. Project Zero describes its reasons as the effect of large language models on finding and exploiting bugs; those reasons are asserted in the post and are set below beside what the saved pages print.

The old rule and the new rule, as each page prints them

The post says Project Zero has followed a 90 + 30 policy since 2021 and replaces it with "Full X": vulnerabilities are disclosed X days after they are reported to the vendor. It says that starting today the policy is Full 90, that on 2 February 2027 Project Zero will move to Full 60, and that on 3 January 2028 it will move to Full 30. It removes most special cases, including the grace period and the variant policy, and keeps three exceptions: bugs actively exploited in the wild, bugs whose details others have already made public, and vendor requests. The table sets it beside the 2021 edition post.

The 2021 rule and the 2026 rule, item by item. Sources: the 2021 edition post and the 2026 post, both as read on 16 November 2026. Nothing in this table is derived.

Item2021 post ("90 + 30")2026 post ("Full X")
Normal deadline90 days. If fixed within 90 days, details 30 days after the fix. If not, details on day 90Details X days after the report, fixed or not: 90 now, 60 from 2 Feb 2027, 30 from 3 Jan 2028
Grace period14 days on request, for a bug the vendor expects to fix within 104 daysRemoved
Variants of an earlier bugAdded to the existing report with no new deadlineVariant policy removed; what replaces it is not stated
Exploited in the wild7 days, with a 3 day grace on a vendor's requestHandled under Google's application security vulnerability disclosure policy, which this site did not read
Details already publicNot stated in the 2021 postProject Zero may publish at its discretion
Vendor requestEarlier disclosure with mutual agreementDetails made public if the vendor requests it
Is the bug fixed?Decides the date: fix plus 30 days if fixed within 90Still tracked, but "will not be a decision point"

The name Full 90 is not new. The 2021 post's own table calls the 2020 trial "Full 90": public disclosure 90 days after the report whatever the fix, with a 14 day grace period. The 2026 policy therefore returns to the 2020 trial's rule without the grace period, and shortens it in two steps. The 2021 post also said it was likely that Project Zero could move to an "84+28" model for 2022. The 2026 post describes 90 + 30 as the policy since 2021, and no saved page says why 2022 did not bring that change.

Drawn from the 2021 edition post, the 2026 post, the standing policy page and the FAQ, as read on 16 November 2026. Spans and first deadlines marked derived are this site's arithmetic.

When the new clock is shorter than 90 + 30, and when it is not

The table gives the day on which details go public for a bug fixed on a given day, under the 2021 rule (fix plus 30 days, no grace taken) and under each step of Full X. The Full X columns do not depend on the fix.

Day of disclosure by day of fix. Derived: the 2021 column is the fix day plus 30, from the 2021 post; the Full X columns are the standing policy page's rule (its example: a fix on day 47 is disclosed on day 90).

Fixed on day2021 rule: details on dayFull 90: dayFull 60, then Full 30: day
0309060, then 30
20509060, then 30
30609060, then 30
47779060, then 30
59899060, then 30
60909060, then 30
751059060, then 30
901209060, then 30
Not fixed90, or later if a 14 day grace was granted9060, then 30

The break even points follow (derived). Full 90 publishes sooner than the 2021 rule only for fixes after day 60. Full 60 does so only for fixes after day 30. Full 30 does so for every fix after day 0, and is never later. The old worst case for a fixed bug was day 120 (a fix on day 90, plus 30), so Full 30 is 90 days shorter at that end, a quarter of the old span (derived: 30 / 120). Applied to the Reporting Transparency table by Fix time, 34 of the 61 closed rows were fixed before day 60, where the 2021 rule would have published sooner than Full 90 would; 3 sit at exactly 60 days, a tie; and 20 fall between 61 and 90 days, where Full 90 is sooner. The other 4 are above 90 days, where the 2021 outcome depends on whether a grace period was granted, which the table does not say. This is a counterfactual, not a record: no saved page prints the date on which details were published for any of these rows.

Which clock applies to a given bug is a separate question. The FAQ says "Vulnerabilities we report are subject to the disclosure policy on the date they were reported." This site reads that as: a report made up to 1 February 2027 keeps Full 90 and a report made up to 2 January 2028 keeps Full 60. The FAQ does not apply the rule to the two switch dates, so this is a reading, not a quotation. On that reading the first Full 90 deadline falls on 11 February 2027, the last Full 90 deadline on 2 May 2027, the first Full 60 deadline on 3 April 2027, the last Full 60 deadline on 2 March 2028, and the first Full 30 deadline on 2 February 2028, 446 days after the post's date (all derived: start date plus X days; 416 + 30 = 446). A vendor is first held to 30 days, on this reading, well over a year after the post.

What Project Zero's own table shows, and what it measures

The page has three tables. The first lists open reports inside their deadline, with a column headed "90-day Deadline expires": 49 rows. The second lists closed reports with Date Fixed, Fix time (Days) and First Known User Device Protected: 61 rows. The third lists reports whose deadline expired without a fix: 5 rows. That is 115 rows (derived). The page says it moves a report into the second table once it is marked Fixed, Working as Intended or Won't Fix, and into the third if the deadline expires unfixed. This site checked that the printed Fix time equals Date Fixed minus Date Reported on all 61 rows.

Drawn from the second table of the Reporting Transparency page, most recent update 10 November 2026, as read on 16 November 2026. Every closed row is plotted; none is excluded. Counts over 30, 60 and 90 are this site's (derived).

How the counts were made. All 61 rows of the second table were counted and none excluded. Over 30 days: 42. Over 60 days: 24. Over 90 days: 4 (Samsung Mobile Audio Codecs at 92, Samsung Semiconductor Xclipse at 94, Google Pixel at 95, Imagination PowerVR at 101). At 30 days or fewer: 19. As shares of 61, that is about 69 percent over 30, 39 percent over 60 and 7 percent over 90 (derived). The mean is 49.2 days (derived: 3,000 days over 61 rows). The word Unknown appears on this page in the last column, First Known User Device Protected, not in Date Fixed: every one of the 61 rows prints a fix date and a fix time.

Rows per vendor on the Reporting Transparency page, most recent update 10 November 2026. Vendor names as printed in the Reported to column. All counts are this site's (derived).

Reported toClosed rows (fix time over 30 days)Open, inside the deadlinePast deadline, unfixed
Apple14 (12 over 30 days)300
Adobe9 (6 over 30 days)70
Google6 (3 over 30 days)45
Samsung Mobile9 (9 over 30 days)30
Linux5 (1 over 30 days)00
Oracle5 (5 over 30 days)00
FreeType4 (0 over 30 days)00
Imagination4 (4 over 30 days)00
Cisco020
Ittiam Systems020
Microsoft2 (0 over 30 days)00
Samsung Semiconductor2 (2 over 30 days)00
AMD010
StrukturAG1 (0 over 30 days)00
All 14 names61 (42 over 30 days)495

The open rows. Of the 49 open rows, 30 are Apple's, 7 Adobe's, 4 Google's, 3 Samsung Mobile's, 2 Ittiam Systems', 2 Cisco's and 1 AMD's (derived). On the page's update date, 22 of the 49 were already more than 30 days old and 12 more than 60 (derived: 10 November minus Date Reported). Those 12 are all Apple Image Codecs rows, 8 with a deadline of 23 November 2026 and 4 with 25 November, 7 and 9 days after 16 November (derived). They were reported under the old rule, and the FAQ ties each report to the policy of its report date, so this site says nothing about when their details will appear.

The rows past their deadline. All 5 rows in the third table are Google's: four Pixel and one Android, reported between 17 March and 24 June 2026, and 139 to 238 days old on the page's update date (derived). The page does not say whether details of these five are public. Each row links to the issue tracker; this site did not open those entries.

The patch gap column. The post asks vendors to reduce patch gap, the time between a fix being released and the user installing it. The nearest thing on the table is the column First Known User Device Protected. It prints Unknown on 50 of the 61 closed rows (derived), and the 11 rows that print a date (5 Oracle, 3 Google, 2 Microsoft, 1 Apple) all print the same date as Date Fixed (derived: compared row by row). The page does not define the column beyond its heading, so this site draws no patch gap figure from it.

What the post asserts, and what a saved page measures

The post gives three reasons for the change, all about large language models, and a fourth point about vendor feedback. Each is Project Zero's own statement. The table sets each beside what a saved page prints about it.

Claims in the 2026 post against what the saved pages print. Sources: the 2026 post and the Reporting Transparency page as read on 16 November 2026. The weekly rate is derived: 115 rows over 303 days, times 7.

Claim in the postWhat a saved page printsNot stated
Large language models raise the chance that several parties, including malicious ones, find the same bug during the embargoThe claim, as an assertionAny count of bugs found twice, or the time to a second discovery
Zero patch-to-exploitation time is "a reasonable, though imperfect" model of attacker behaviourThe claim, offered as a model; the post says patch soak time is no longer meaningfulAny measured time from a patch to a working exploit
Project Zero finds several vulnerabilities a week where it once found a few a yearThe Transparency page lists 115 rows dated 7 January to 6 November 2026, about 2.7 rows a week (derived)A weekly count, or whether the page lists every report; so this is not a test of the claim
Vendors asked for a gradual pathThe post says Project Zero asked the vendors it reports most to, and lists three concerns: unfixed bugs becoming available to attackers, fixing on deadline ahead of more severe bugs, and vendors giving up if they cannot meet the deadlineWhich vendors, and any vendor statement
Severity is not an effective metric for disclosure decisionsThe same post says Project Zero reports only bugs with substantial severity and demonstrated security impactHow severity is judged, or how many reports a vendor treats as low quality
Drawn from the 2026 post and the Reporting Transparency page as read on 16 November 2026. The count of 115 rows is this site's (derived).

Where the pages differ

The pages disagree or leave gaps in several places. The table prints what each says and resolves none of them.

Differences between the saved pages, as read on 16 November 2026. This site resolves none of them. Derived values are labelled.

PointOne page printsAnother page prints
Date of the changeThe post's own date link reads 2026-Nov-13, and the text says "Starting today"Help Net Security, a relay read for its date only: published 16 November 2026 at 06:30 UTC. Both dates are printed here as they are
In force, or plannedPost: starting today the policy is Full 90, and Project Zero "will move" to Full 60 and Full 30Policy page: X "is currently 90" and "we plan to move" to 60 and 30. The FAQ also says plan
The Transparency pageStill describes itself as part of the 2025 Policy Trial, still carries a 90-day deadline column, and says links to Fixed issues stay private for up to 30 days after a fixThe post says patch soak time is no longer meaningful. This site has not seen the page updated for Full X: its update is dated 10 Nov, 3 days before the post (derived)
Whether a bug is fixedPost: "We still plan to track whether bugs are fixed", but this "will not be a decision point"The Transparency page still sorts rows by Fixed, Working as Intended or Won't Fix and prints fix times
SeverityPost: severity is not an effective metric for making disclosure decisionsSame post: Project Zero only reports bugs "with substantial severity and demonstrated security impact"
Exploited in the wildPolicy page: disclosed after 7 days, under Google's own policy. 2021 post: a 3 day grace on request. The FAQ lists four in-the-wild cases extended by 23, 101, 98 and 98 days under a 7 day deadlinePost: the grace period is removed. What that does to the in-the-wild case is not stated on any saved page; Google's policy page was not read
Details already publicPost: Project Zero "may" publish at its discretionPolicy page: Project Zero "will release" details immediately if another party publishes them
How many exceptionsPolicy page: exceptions granted "five times" over 12 yearsFAQ: 10 of 2219 cases extended (0.45 percent, derived). The 10 sit under five write-ups (derived grouping; no page says that is how five arises). Post: exceptions on "rare occasions", including Spectre and Meltdown, "and will continue to do so"
HardwarePost: the section headed hardware asks vendors to plan for imminent disclosure and says exceptions were granted before, including Spectre and MeltdownHelp Net: "Hardware bugs may still get extra time." The post does not name hardware as a category of exception
How often the table updatesTransparency page: weekly. FAQ: within one weekPolicy page: soon after. The latest report in the table is dated 6 Nov and the update 10 Nov, 4 days later (derived)
The 2022 plan2021 post: likely a move to "84+28" for 20222026 post: 90 + 30 since 2021. No saved page explains the gap

One company on both sides of the deadline

Project Zero is a Google team, and Google is also among the vendors it reports to. Google appears in 15 of the 115 rows (4 open, 6 closed, 5 past deadline; derived), under product names including Pixel, Android, V8, Brotli and libgav1. The FAQ says that when Google is the recipient Project Zero follows Google's standard external bug reporting procedures, the same processes that a non Google researcher would meet, and that Project Zero does not share reports internally. All 5 rows past their deadline are Google rows. The saved pages do not say why, and they do not say whether Google's own products are held to the same deadline beyond the policy text. Bugs exploited in the wild go to a Google policy page that this site did not read.

What the pages state, and what they do not

What the saved pages state and do not state, as read on 16 November 2026. Sources: the 2026 post, the standing policy page, the FAQ, the Reporting Transparency page and the 2021 post.

TopicStated (by which page)Not stated
The changeFull X: details X days after the report, fixed or not; 90, then 60 from 2 Feb 2027, then 30 from 3 Jan 2028 (post, policy page, FAQ)Any end date for Full 30
Vendors affectedThe post says it asked the vendors it reports most to; the table names 14 reported-to names (derived count)How many vendors are affected, and which ones object, agree or have changed a process
Evidence for the reasonsAssertions in the postA measurement of patch-to-exploit time; a count of duplicate discoveries; a count of reports per week or per year
Harm from an unfixed bug at the deadlineThe post reports vendors' concern that an immediate change could harm users, and stages the cutAny case or count of such harm
Can vendors meet 60 or 30 daysThe table: 19 of 61 closed rows at 30 days or fewer, under the old rule (derived)Any data under Full X; any vendor commitment
Hardware and other exceptionsExceptions granted before, including Spectre and Meltdown, and to be granted again; five exceptions in 12 years (policy page)Which future cases, or how hardware is treated as a class
Who else will followNothingAny other team or programme
Technical artifactsNone. The pages print a policy, not indicators, code or configuration, and none is reproduced hereAnything a defender could hunt for

What the post asks of vendors, and what it links to

The post asks vendors to move towards remediating vulnerabilities faster, to reduce patch gap by shortening rollout or by other protection while a fix is in progress, and to consider how they would answer an imminent disclosure of a serious bug, including non patch strategies, before one arrives. The linked post "How to fix a bug in a fix", dated 2026-Oct-06 on its page, covers what that means in practice. It lists six stages of patching (triage, patch development, testing, partner review, delivery and activation), says testing and delivery limit emergency patch speed more than triage or development, and describes four methods large vendors use to act faster than their usual update: feature flags, filtering, alternate update channels and hotpatching, with the risks it sees in each (for example, libraries delivered outside normal updates must be verified, and hotpatching can introduce ways to bypass exploit mitigations). That post's closing section calls it urgent for vendors to plan for the worst case of widespread active exploitation.

What to do, in order

The first item is Project Zero's own ask. The others are this site's suggestions and are labelled as such.

Take this with you

Actions, in order

  • Software vendors, Project Zero's own ask: shorten remediation time and patch gap, and plan non patch protection and emergency patching before a report arrives. The linked post lists feature flags, filtering, alternate channels and hotpatching as systems large vendors use.
  • Teams that ship software from a vendor Project Zero reports to, this site's suggestion: measure how long your own deployment of a vendor fix takes, from the vendor's release to the last affected system, and compare it with 30 days. Details can now appear on day 90 with no fix, then on day 60 and day 30.
  • The same teams, this site's suggestion: record the security contact and the advisory channel for each vendor you depend on. The Transparency page lists 14 reported-to names, from Apple and Adobe to FreeType and Linux.
  • Security teams reading advisories, this site's suggestion: treat a Project Zero write-up published at its deadline for a bug the vendor has not fixed as a prompt to plan a mitigation, not to wait for a patch. The post itself says non patch strategies exist.
  • Maintainers of open source or small team software, this site's suggestion: thirty days is a staffing question. Decide who answers a report and within what time before one arrives; the FAQ says Project Zero reports to the vendor or open source project responsible for the fix.

The question the table leaves open

Project Zero describes the schedule as "challenging but realistic". The only table of vendor speed in the pages this site read measures the old rule, lists 42 of its 61 closed reports above 30 days, and shows 5 Google reports past their deadline with no fix (all derived counts from the page's own rows). Where is the evidence that vendors can meet 30 days, when the one table Project Zero publishes has no row reported under the new rule?

Sources

  1. PrimaryPolicy and Disclosure: 2026 Edition, page date 2026-Nov-13: Full X, the 90, 60 and 30 schedule, removed grace period and variant policy, the three exceptions, vendor feedback and the three reasons. Read by this site on 16 November 2026Project Zero (Google)accessed 2026-11-16
  2. PrimaryStanding Vulnerability Disclosure Policy page as read on 16 November 2026: X is currently 90, plans to move to 60 and 30, examples of a fix on day 47, 7 day in the wild rule, five exceptions in 12 yearsProject Zero (Google)accessed 2026-11-16
  3. PrimaryVulnerability Disclosure FAQ as read on 16 November 2026: report-date rule, ten extended cases out of 2219, how Google is treated as a recipient, weekly transparencyProject Zero (Google)accessed 2026-11-16
  4. PrimaryReporting Transparency page, most recent update 10 November 2026: three tables with 49 open, 61 closed and 5 past-deadline rows; the only vendor speed data readProject Zero (Google)accessed 2026-11-16
  5. PrimaryHow to fix a bug in a fix, page date 2026-Oct-06: the stages of patching and four emergency methods, linked from the 2026 postProject Zero (Google)accessed 2026-11-16
  6. PrimaryPolicy and Disclosure: 2021 Edition, page date 2021-Apr-15: the 90 + 30 rule, the 2020 Full 90 trial, the 14 day grace period and the 84+28 remarkProject Zero (Google)accessed 2026-11-16
  7. Reported byRelay of the story, published 16 November 2026 at 06:30 UTC per page data; read for its date and its wording of the hardware point only. SecondaryHelp Net Securityaccessed 2026-11-16

Share this briefing

Know someone who owns this problem? Send it to them.

Related briefings

The briefing, in your inbox

Practitioner analysis of cyber and AI security news. No vendor noise.

How often

Every new briefing in one email, at 7am, or at 7am, 12:30pm and 6pm. Nothing is sent when nothing is new. Unsubscribe any time.