The disclosure window is shrinking: Google Project Zero and Anthropic cut their deadlines within four days
For about fifteen years the number to remember was 90: the days a vendor got between the private report of a vulnerability and its publication. Google Project Zero made it the standard, and in 2021 added a 30-day grace period after the patch. On November 13, 2026, the same team published “Policy and Disclosure: 2026 Edition”, signed by Natalie Silvanovich according to the official post. The document introduces a “Full X” rule: the details of a flaw become public X days after it is reported, whether or not it has been fixed. The schedule sets “Full 90” from November 13, 2026, “Full 60” from February 2, 2027 and “Full 30” from January 3, 2028. Project Zero says the change is gradual because vendors feared that a sudden switch would leave unfixed bugs exposed.
The new policy drops almost every special case, including the grace period of the “90+30” model and the rule on variants. Three exceptions remain: actively exploited flaws, flaws already made public by others, and vendor requests. For the first, the post gives no window in days and points to Google's application security disclosure policy. Project Zero recalls the exceptions it granted for hardware in the past, such as Spectre and Meltdown, and says it will keep granting them. The rationale is the team's own account of its work: according to Project Zero, language models have made attackers faster at finding and exploiting flaws. The post adds that the team now finds several vulnerabilities a week in targets where it used to find only a few a year. That makes it more likely that several parties, malicious ones included, discover the same bug, so waiting after the patch would no longer make sense.
On November 16, Anthropic's Frontier Red Team announced shorter windows for vulnerabilities found with Claude. High and critical severity flaws go from 90 to 30 days, and all others to 45. The new deadlines apply to reports made from January 1, 2027; reports already open on that date keep the old ones. If a maintainer does not respond, Anthropic will turn to an outside coordinator after 15 days instead of 30. After the patch, full technical details will come out after 30 days instead of 45. Actively exploited flaws keep the seven-day window. For software that is hard to update, such as critical infrastructure, or for fixes that require changes to specifications or RFCs, timelines will be agreed with the vendor. High or critical flaws reported more than 90 days ago and still undisclosed will get another 30 days from notification, then go on the company's dashboard. AI-generated findings that no person has reviewed yet, such as those from OSS Scanner, are excluded. Here too the rationale is the company's: a 90-day window, the team writes, made sense when a flaw could stay private for months. Today, it adds, models help develop fixes, and a bug found by one model is more and more likely to be found by another.
Some questions remain open, and they should be kept separate. The first is Anthropic's schedule: the general page of its policy, last updated March 6, 2026, still says 90 days, and it is unclear whether it will be corrected before January. The second is independent confirmation: among the sources consulted, it comes only from DigitalToday, which in turn cites IT Home. I found no coverage in the major security outlets consulted (SecurityWeek, BleepingComputer), and The Record could not be accessed. The third is the rationale: in both cases, the speed-up attributed to models is a claim the two teams make about their own work and that of attackers. The calendar dates are facts; the causes are claims that need checking on their own. For Project Zero, the only source is the official post, and reactions from vendors and other research programs are not documented. Finally, neither post names the other company, so any suggestion of coordination would be an unsourced hypothesis.
It is striking that two of the organizations that hunt vulnerabilities hardest shortened the same wait within four days of each other. Their reasoning is that secrets keep for less time than they used to. According to Project Zero, vendors were already worried about an abrupt change. How they will react now, the sources don't say yet.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I read Project Zero's official post of 11/13/2026 (author, the 90/60/30 schedule and its dates, exceptions, rationale) and the post by Anthropic's Frontier Red Team of 11/16/2026 (30 and 45 days from 1/1/2027, escalation after 15 days, 30 days after the patch, seven days for exploited flaws, exclusions). DigitalToday (11/17/2026) confirms Anthropic's change with the same figures. Anthropic's policy page has not been updated yet. I checked the 90→60→30 schedule reported by an aggregator against Project Zero's original text. I left out an aggregator's claim that Anthropic had 'adopted the same deadline' as Google, because the rules are different.
- Incertezze
- Anthropic's general policy page, last updated March 6, 2026, still says 90 days, and it is unclear whether it will change before January 1, 2027. SecurityWeek and BleepingComputer have not covered it, and The Record could not be accessed. The only independent confirmation of Anthropic's change is DigitalToday, which cites IT Home. For Project Zero, the only source is the official post, which sets no window in days for actively exploited flaws and points to Google's policy. Reactions from vendors and other research programs are not documented.
- Perché pubblicarla
- Within days of each other, two of the leading vulnerability hunters are dropping the 90-day rule, in place for more than a decade, and both name AI as the cause. Software vendors, open source maintainers and IT managers will have to fix flaws much faster.