AI is changing the time defenders have to fix a vulnerability
Software security has long relied on a coordinated process. A researcher discovers a vulnerability, reports it privately to the software maintainer, gives the maintainer time to prepare a fix and then publishes technical details so users and the wider security community can understand the issue.
That process depends on time.
On November 16, 2026, Anthropic announced that it is shortening the timelines in its coordinated vulnerability disclosure policy. The company says its frontier models have become better at discovering vulnerabilities, and that the possibility of another researcher or model independently finding the same flaw has increased.
Under the revised policy, Anthropic will move from a 90 day disclosure window to:
- 30 days for high and critical severity vulnerabilities.
- 45 days for all other vulnerabilities.
The new timelines apply to vulnerabilities reported to vendors and maintainers on or after January 1, 2027. Reports already in progress on that date retain their existing deadlines.
This is not simply a policy housekeeping exercise. It reflects a difficult security question: if AI can help discover software flaws faster, should the industry also shorten the time available to fix them?
What exactly is changing?
Anthropic’s November 16 update describes several changes to its coordinated vulnerability disclosure process.
| Policy area | Earlier timeline | Revised timeline |
|---|---|---|
| High and critical severity disclosure | 90 days | 30 days |
| Other vulnerability disclosure | 90 days | 45 days |
| Escalation to an external coordinator if a maintainer does not respond | 30 days | 15 days |
| Publication of full technical details after a patch is available | 45 days | 30 days |
The seven day disclosure window for vulnerabilities under active exploitation remains unchanged.
Anthropic also says that some vulnerabilities cannot realistically be fixed within 30 or 45 days. Examples include flaws in difficult to deploy critical infrastructure software and issues that require changes to a technical specification or RFC. In those cases, it says it will work with the vendor on a remediation timeline.
The policy does not apply to AI-discovered findings that have not received human review, such as findings from its OSS Scanner.
These qualifications matter. The change is not a blanket instruction that every software vulnerability must be publicly disclosed after exactly 30 days.
Why does Anthropic believe AI changes the disclosure process?
Traditional disclosure timelines were designed around a world in which vulnerability discovery and remediation moved largely at human speed.
A researcher might find a flaw, document it and privately notify a maintainer. The maintainer would then need time to understand the issue, reproduce it, develop a patch, test the patch and coordinate a release.
Anthropic argues that AI changes two parts of this equation.
First, models can help security researchers discover vulnerabilities more quickly.
Second, a vulnerability found by one model may be found independently by another model or researcher. The period during which a flaw remains unknown to attackers may therefore become less predictable.
Anthropic says it is seeing a higher rate of vulnerability rediscovery as model capabilities improve.
The company’s argument is that a 90 day window may no longer provide the same balance between responsible remediation and public disclosure that it once did.
This is Anthropic’s rationale for changing its own policy. It should not be interpreted as proof that every AI system can reliably discover every type of vulnerability or that all software maintainers can safely patch complex flaws within a month.
The security trade-off: faster disclosure versus safe remediation
Shorter disclosure windows can help users learn about vulnerabilities sooner and encourage faster fixes.
But they also create pressure.
A maintainer of a widely used open source library may be responsible for code that runs in thousands of organisations. A rushed patch can introduce new bugs or break compatibility. Some projects are maintained by small teams or volunteers who do not have a dedicated security department.
A shorter deadline does not automatically create more engineering capacity.
This is why Anthropic says it will continue to provide actionable guidance and, where possible, a patch or mitigation alongside vulnerability reports. It also says it will agree on a sustainable pace with maintainers before sending large volumes of findings to a single project.
The exception process for difficult to fix software is another important part of the policy.
The challenge is to reduce the window of exposure without encouraging unsafe fixes or overwhelming the people responsible for maintaining critical software.
What happens to open source maintainers?
Open source software is used throughout the modern technology stack. A vulnerability in a widely adopted library can affect cloud services, enterprise applications, developer tools and consumer products.
Many maintainers do not have the resources of a large technology company.
Anthropic says it is continuing programmes intended to support maintainers and defenders:
- OSS Scanner: Eligible maintainers of critical open source projects can apply for a no cost service that periodically scans code with Anthropic’s strongest models and sends human reviewed findings.
- Claude for Open Source Program: Eligible maintainers and contributors can apply for six months of Claude Ultra at no cost.
- Cyber Verification Program: Security professionals, including open source maintainers, can apply for safeguards tailored to legitimate defensive work.
These programmes may help some teams, but they do not remove the need for maintainers to triage reports, validate findings, test fixes and communicate with users.
What should Indian technology companies do?
Indian software companies, SaaS providers, fintech businesses, IT service providers and open source maintainers should treat the policy change as a prompt to review their own vulnerability response processes.
A practical checklist:
1. Publish a clear security contact
Make it easy for researchers to report vulnerabilities. Provide a security policy and a monitored reporting channel.
2. Define severity and response ownership
Identify who receives a report, who validates it, who assigns severity and who approves a patch.
3. Prepare a rapid triage process
Create a way to reproduce findings, assess affected versions, identify exposure and decide whether a temporary mitigation is needed.
4. Test patches before release
Fast remediation should not mean skipping regression tests, compatibility checks or deployment planning.
5. Communicate with customers
Prepare a process for security advisories, affected version lists, workarounds and patch availability.
6. Use AI as an assistant, not as the final security authority
AI tools can help analyse code, explain findings and suggest fixes. Human security engineers should validate the vulnerability and the remediation, especially for high impact systems.
A proposed vulnerability response readiness score
The following is an editorial assessment framework, not an Anthropic standard.
| Capability | Weight |
|---|---|
| Security reporting and ownership | 15% |
| Triage and severity assessment | 20% |
| Reproduction and validation | 15% |
| Patch development and testing | 25% |
| Customer communication and disclosure | 15% |
| Maintainer capacity and escalation | 10% |
| Total | 100% |
Score each area from zero to five and calculate the weighted total. The score can help an organisation identify process gaps before a serious vulnerability arrives.
A high score is not a guarantee of security. It indicates that the organisation has a more organised response process.
What does this mean for the future of AI security?
AI is creating a cycle in which the same technology can help defenders and attackers.
Models can assist with code review, vulnerability discovery, exploit analysis and patch development. Those capabilities can help security teams work faster. They can also make it easier for more actors to search for weaknesses.
The response cannot be limited to buying another security tool.
Organisations need a process that connects discovery, validation, remediation, testing and communication. Open source maintainers need support, not just shorter deadlines.
Anthropic’s policy change is one company’s response to this changing environment. Other vendors and security organisations may choose different timelines based on their own risk models and operational experience.
The larger lesson is that cybersecurity processes designed for a slower discovery cycle may need to evolve as AI capabilities improve.
AI may shorten the time needed to find a vulnerability. The harder challenge is making sure the people responsible for fixing it can move just as quickly without compromising safety.
Sources and further reading
- Anthropic: Updating our coordinated vulnerability disclosure policy timelines
- Anthropic: Coordinated vulnerability disclosure dashboard
Editorial note: Policy dates, timelines and programme descriptions are based on Anthropic’s published policy update. The implications for maintainers and Indian technology companies are analysis, not claims that all organisations face identical timelines.