Anthropic vulnerability disclosure: The 30-day shift13 min read2,513 words

Anthropic vulnerability disclosure: The 30-day shift

Anthropic will disclose high- and critical-severity flaws after 30 days from 2027. See what changes for maintainers, security teams and software buyers.

Anthropic vulnerability disclosureAI security vulnerabilitiescoordinated vulnerability disclosuresoftware security patch deadlinesFrontier Red Team
Anthropic vulnerability disclosure: The 30-day shift

Why Anthropic Is Cutting Its Vulnerability Disclosure Window to 30 Days

A vulnerability disclosure deadline is a balancing act: give a software team time to investigate and fix a problem, but don't leave users guessing forever. Anthropic is tightening that window. Starting in 2027, its Frontier Red Team will publish high- and critical-severity vulnerability reports 30 days after it notifies a maintainer (down from 90 days). Other flaws will get a 45-day window.

The underlying reason is pretty straightforward: if AI makes it easier to find the same weakness, more researchers and tools may surface the issue independently. A bug that once might have stayed quiet while a vendor worked on a patch can become public faster than anyone expected. That doesn't automatically mean every AI-generated lead is correct - or that a serious fix can always be completed in a month. But it does make the time between discovery and disclosure a more consequential decision.

The updated deadlines apply to reports made on or after January 1, 2027. Existing reports keep their current timelines. For certain older, unpublished serious flaws, Anthropic is offering a separate 30-day extension. Below is what's changing, why it matters, and what maintainers and organizations can do to prepare - without treating a shorter deadline as a shortcut around good security practices.

What is changing in Anthropic's disclosure policy?

Anthropic's revised policy shortens parts of its vulnerability disclosure process. The headline change is the publication deadline: for high- and critical-severity flaws, Anthropic will disclose reports 30 days after notifying the maintainer instead of waiting 90 days. For other flaws, the publication deadline becomes 45 days.

These timelines describe when Anthropic may publish its findings. They do not promise that a vendor can develop, review, and deploy a fix within that same period. In practice, what it takes to remediate depends on the software, how releases are managed, and the risk of introducing a regression.

The new timelines at a glance

Situation Anthropic's stated timeline
High- or critical-severity flaw Public disclosure 30 days after reporting
Other flaw Public disclosure after 45 days
No response to the initial report Escalation to an outside coordinator after 15 days, down from 30
Full technical details after a patch is released 30 days, down from 45
Flaw under active attack Seven-day window remains unchanged

The policy applies to reports made on or after January 1, 2027. Reports already in progress keep their existing deadlines. Anthropic also says it will give maintainers of certain serious flaws reported more than 90 days ago an additional 30 days before those reports are published under the new rules.

There are exceptions. Some vulnerabilities - such as issues in critical infrastructure software or in a technical standard - may require longer. In those cases, Anthropic says it will agree on a timeline with the vendor. The seven-day window for flaws under active attack stays as it is.

Anthropic also excludes AI findings that haven't been reviewed by a person, including results from its OSS Scanner service. In other words, a model output isn't automatically treated as a validated vulnerability report here; human review stays part of the process.

What the shorter deadlines do not mean

A 30-day policy doesn't mean Anthropic will always publish exactly 30 days after discovery, or that every vendor gets a clean 30 days to finish every type of remediation. The clock starts when the report goes to the maintainer, and the policy allows for longer schedules in certain difficult situations.

It also doesn't mean a vulnerability is automatically "safe" once a patch exists. Anthropic says it will wait 30 days after a patch is released before sharing the full technical details. The point is to give users and operators time to adopt the fix before details that could help reproduce the flaw are published.

Why does Anthropic think 90 days is no longer enough?

Anthropic's argument comes down to speed - how quickly vulnerabilities can be found and how likely it is that multiple people will discover the same weakness independently. Its earlier approach assumed a flaw could stay private long enough for a maintainer to work on a fix. The company says that assumption is less reliable as AI tools get better at spotting software weaknesses.

The issue isn't just that a model can find a bug. It's that the same weakness may be identified by more than one researcher or system. If discovery becomes easier to repeat, a long confidential period may provide less protection than it used to. A vulnerability can remain unknown to the public and still be at risk of being independently discovered.

Faster discovery changes the disclosure trade-off

Coordinated disclosure has two goals pulling in different directions. Vendors need time to understand a report, develop a fix, and distribute it. Users need enough information to respond to the risk - especially if patches are delayed or if the weakness becomes known outside the reporting process.

A 90-day window gives vendors more time. A 30-day window shifts toward faster public awareness. Either way, the trade-off doesn't disappear. The real question is whether extra time for remediation is worth the added risk of leaving the flaw unpatched and undisclosed for longer.

That balance varies by software. A small project with a maintainer who can quickly review and ship a targeted fix might be able to respond quickly. A widely used component - deployed across banks, hospitals, cloud providers, or public infrastructure - often requires coordination across teams and release cycles. A fixed deadline can't make those operating realities uniform.

Anthropic says its policy accounts for this by allowing agreed timelines for complex cases. That flexibility still depends on communication: maintainers need to respond early, explain constraints, and keep the reporting team in the loop. If nobody responds, escalation becomes more likely - not less.

AI increases both the speed and the uncertainty

AI-assisted security work shouldn't be treated like a one-button capability. Tools can help researchers inspect code and spot suspicious patterns, but a suspected issue still has to be validated. Some "findings" can be false positives, hard to reproduce, or highly dependent on a particular configuration.

That's one reason Anthropic draws a line between automated output and a report that enters its disclosure policy: findings must be reviewed by a person. For maintainers, the operational challenge is verifying quickly enough to make decisions - without assuming every machine-generated alert has the same severity or real-world impact.

Overall, the policy responds to a real shift in the discovery environment, but it doesn't claim there's a single disclosure speed that fits every vulnerability. How well it works will depend on report quality, how quickly human validation happens, and whether exceptions are handled consistently.

What does the shorter window mean for maintainers and users?

For maintainers, the biggest change isn't only the shorter publication deadline. It's the need to make triage and communication work under a tighter timeline. If a report sits unattended for two weeks, it burns a large chunk of a 30-day window before anyone even assesses severity.

Anthropic says it will refer a report to an outside coordinator after 15 days without a response, down from 30. That makes clear security contacts and routing processes more important. A small team may not need a full security department, but it does need a way to confirm receipt and identify who is responsible for investigating.

A workable response process for maintainers

A simple process can cut avoidable delays without promising a fix before anyone understands the issue:

  1. Acknowledge the report. Confirm the report reached the right person and share a channel for follow-up.
  2. Validate the finding. Reproduce what's claimed, check affected versions, and separate confirmed impact from assumptions.
  3. Set an owner and timeline. Assign responsibility for investigation, remediation, testing, and release coordination.
  4. Communicate constraints early. If a fix requires a major release or coordination with downstream users, explain that before the deadline starts to feel urgent.
  5. Prepare a safe remediation path. Where possible, develop and test a patch, then give users a clear way to apply it.

Anthropic says it will continue to provide fix guidance and, when possible, a patch with each report. That help may reduce some engineering burden, but maintainers still need to review changes in the context of their own codebase. Applying an unreviewed patch under pressure can create a second production problem while trying to solve the first.

What software buyers and security teams should do

If your organization relies on open-source or third-party software, don't assume a vendor's disclosure timeline matches your own internal response timeline. A published report can trigger work across asset inventory, risk assessment, testing, and deployment. If you don't know where a component is used, even a timely patch may not reduce exposure quickly.

Security and engineering teams can use this policy change as an excuse to shore up a few operational basics:

  • Can you identify where critical dependencies are deployed?
  • Who receives vulnerability reports, and who decides whether they matter in your environment?
  • Can you test and roll out a security update outside the normal release cycle?
  • Do suppliers and internal teams have a way to communicate when a patch isn't immediately available?

A shorter window may give users faster access to actionable information. It can also mean more visible vulnerabilities before fixes are complete or upgrade paths are straightforward. The practical response isn't to wait for disclosure - it's to make dependency tracking and patch handling part of routine operations.

How does Anthropic's policy compare with other disclosure approaches?

Anthropic isn't the only organization moving toward shorter disclosure timelines. SecurityWeek reports that Google's Project Zero is also shrinking its window. It plans to move to 60 days starting February 2, 2027, and then down to 30 days starting January 3, 2028.

That comparison is useful, but it shouldn't be treated as proof of a universal standard arriving overnight. The schedules aren't identical, and disclosure policies differ in what they cover, how severity is assessed, and what they do when vendors need more time. A number alone doesn't tell you whether a policy will work for a particular vulnerability or software ecosystem.

Compare the process, not just the number

When evaluating a disclosure policy, it's worth looking beyond the headline deadline. What really determines how it plays out includes:

  • When the clock starts: A deadline measured from discovery is different from one measured from notification to the maintainer.
  • How severity affects timing: Policies often treat high-impact flaws differently from lower-risk findings.
  • What happens when a maintainer is unresponsive: Escalation rules can affect whether a report gets attention before publication.
  • Whether exceptions are available: Some issues require coordination across vendors, infrastructure operators, or standards bodies.
  • How long users get after a patch: Publishing full technical details can create a separate update window.
  • Whether findings are validated: Human review helps separate a confirmed vulnerability from an unverified automated result.

A disclosure timeline is a system, not a single countdown. A short deadline without a workable escalation path can leave a report ignored. A longer deadline without clear expectations can leave users exposed without a good plan. In the end, policy quality comes down to how reporting, triage, remediation, and publication fit together.

For software teams, the takeaway isn't to optimize for one external deadline. Build a process that can handle reports from different researchers and vendors while keeping enough flexibility for issues that can't be safely fixed on a standard schedule. For organizations buying software, ask how suppliers notify customers and deliver fixes - not only how quickly they publish vulnerability information.

Conclusion: treat disclosure time as an operational constraint

Anthropic's move from 90 days to 30 for high- and critical-severity flaws reflects a changed assumption: vulnerabilities may be harder to keep private as AI-assisted discovery improves. Shorter timelines can give users earlier visibility, but they also squeeze the time maintainers need to verify reports, coordinate fixes, and ship safe updates.

The practical response is preparation, not panic. Maintainers should make it easy for reports to land and get answered, while organizations should know which dependencies they use and how quickly they can deploy a fix. A disclosure deadline can create urgency; only a functioning response process turns that urgency into reduced risk.

Frequently Asked Questions

The revised deadlines apply to vulnerability reports made on or after January 1, 2027. Reports already in progress remain under their existing deadlines. Anthropic has separately told maintainers of certain serious, older unpublished flaws that they have 30 more days before possible publication.
No. The 30-day period applies to high- and critical-severity flaws. Other flaws have a 45-day disclosure window. Anthropic also says it may agree on a longer timeline with a vendor for certain complex cases, such as issues involving critical infrastructure software or a technical standard.
If a maintainer does not answer Anthropic’s initial report, Anthropic says it will refer the case to an outside coordinator after 15 days. That is down from 30 days under the previous process. The escalation is intended to help move the report forward; it does not itself mean a patch has been produced.
Not when the finding has not been reviewed by a person. Anthropic says the policy does not cover unreviewed AI findings, including those from its OSS Scanner service. That distinction separates automated output from findings that have received human review before entering the disclosure process.

Related articles