Google's bug hunters will publish software flaws 90 days after reporting them, fixed or not, and cut that to 30 days by 2028
Google's Project Zero, one of the most influential bug-hunting teams in the software industry, changed its rules on Friday. It will now publish details of the security flaws it finds 90 days after telling the software maker, whether or not a fix exists.
That window will shrink. It drops to 60 days on February 2, 2027, and to 30 days on January 3, 2028, according to the team's policy post. Project Zero's main stated reason is AI: it says attackers can now find and exploit flaws far faster than the old rules assumed.
For any company that sells software or depends on it, the message is the same. The gap between a flaw being found and the world learning about it is getting shorter, and the clock no longer pauses for slow fixers.
What changes
Since 2021, Project Zero has run a system it called "90 + 30." Software makers had 90 days to fix a flaw. If they did, the team waited another 30 days before publishing details, so customers had time to install the update. Vendors could also ask for a 14-day extension.
That 30-day buffer, which security teams call soak time, is gone. So are the extensions and most special cases. Under the new system, the disclosure date is fixed the day a bug is reported. Whether it has been fixed no longer changes it.
Three exceptions remain. Flaws already being used in attacks fall under a separate Google policy. Details others have already made public can be released early. And a vendor can ask for earlier disclosure. The team says it has occasionally granted exceptions, including for the Spectre and Meltdown chip flaws, and will continue to do so.
A flaw's severity will not change the timeline either. In the post, researcher Natalie Silvanovich argues that severity "does not affect the likelihood" that attackers will discover a bug.
Google's case: AI erased the defenders' head start
The argument rests on two claims about AI. First, AI tools make it more likely that several parties, including criminals, find the same flaw on their own. Second, AI lets attackers study a published fix and quickly build a working attack from it.
The post treats the time between a fix and an attack as effectively zero, calling that "a reasonable, though imperfect, model." If that holds, the 30-day soak time protected no one. A fix reveals the flaw, so attackers could move while customers were still installing it.
Project Zero offers one data point on how much faster discovery has become. It says it now finds "several vulnerabilities a week" in products where it once found a few a year. It gives no fuller figures.
Outside evidence points the same way, with limits. In June, Anthropic's own red team reported that its most capable model turned a public Firefox fix into working attack code in just under an hour. That was a company testing its own product. Google's Mandiant unit estimated in March that attackers, on average, start exploiting flaws seven days before a fix exists. Mandiant did not tie that figure to AI.
What it means for buyers, boards and investors
The change puts the burden on software makers. Project Zero says it asked the vendors it reports the most bugs to for feedback, and their main concern was that an abrupt cut would expose unfixed bugs to attackers. Vendors also worried tight deadlines would skew how they prioritize fixes, or lead some to stop trying to meet deadlines at all.
Those concerns are likely sharpest for companies on slow update cycles, such as hardware makers, which the post addresses separately. Project Zero urges every vendor to plan for "imminent disclosure" of a serious flaw, including ways to protect users that do not depend on a patch.
A recent case shows what public, unfixed flaws can cost. On November 9, N-able, which sells software for managing company computers remotely, took its hosted N-central platform offline across the Americas, Europe and Asia-Pacific. A researcher had published more than 30 alleged flaws with no warning, BleepingComputer reported. The researcher said he found them with AI help and went public after N-able left earlier reports unanswered. N-able said it had seen no active exploitation and later released a fix.
That was a no-notice release, far harsher than Project Zero's 90 days. But it shows the business disruption when attack details arrive before a fix.
For customers, the practical question for suppliers is simple: how fast do you ship fixes, and how fast can we install them? Tech analysis site byteiota argues that companies taking 45 to 60 days to roll out updates will be exposed once the 30-day rule arrives.
The rules matter beyond Google. The 90-day deadline, popularized by teams such as Project Zero, became a rough industry default, a November Cloud Security Alliance paper notes. If other research teams follow again, shorter clocks could become standard, rewarding vendors that can ship fixes in days rather than months.
Two caveats apply. The policy covers only flaws Project Zero itself reports. And Google, which sells security services through Mandiant, both sets this clock and has a commercial stake in how the industry responds. Its AI claims are its own, and it published little data to back them.