Anthropic cuts disclosure deadline for serious vulnerabilities it finds to 30 days
From the start of 2027, Anthropic will give software maintainers 30 days before it discloses a high- or critical-severity vulnerability it has found, and 45 days in other cases. The current limit is 90 days.
Anthropic’s Frontier Red Team, which tests AI capabilities, announced the change on the company’s website on 16 November. The change concerns the company’s own coordinated vulnerability disclosure policy. Under such a policy, whoever finds a vulnerability reports it privately to the software vendor or maintainer first and discloses it publicly only after a deadline, so there is time to fix it.
The policy covers vulnerabilities that Anthropic finds in other organisations’ software. The new deadlines apply to reports made on or after 1 January 2027. Cases already in progress keep their existing deadlines.
Two other time limits are also getting shorter. If a maintainer does not respond to a report, Anthropic will aim to escalate the matter to an external vulnerability coordinator after 15 days. The current limit is 30 days. Once a patch is available, the company will publish full technical details after 30 days instead of the current 45.
Older cases have their own arrangement. Some high- and critical-severity vulnerabilities that Anthropic reported more than 90 days ago remain undisclosed, even though the current deadline has passed. The company is giving their maintainers an additional 30 days. After that, it will publish the reports on its dashboard under the new policy from the start of 2027.
Anthropic says the same bugs are increasingly being rediscovered
Anthropic said disclosure practices rested on two assumptions: fixes move at human pace, and someone else is unlikely to find the same bug independently. Neither assumption holds any longer, the company said. Models can help develop fixes. If one model finds a bug, another is increasingly likely to find it too. The company said it has seen the same vulnerabilities rediscovered more often than before. It gave no figures.
Exceptions and help for maintainers
Not every bug can be fixed in 30 or 45 days. Anthropic will agree a fix timeline with the vendor if deployed systems are hard to update, as in critical infrastructure. It will do the same if a fix requires a change to a specification or an RFC standard. The seven-day deadline for vulnerabilities under active exploitation remains unchanged. The new policy does not apply to AI findings that no human has reviewed, such as those from the OSS Scanner service.
Anthropic acknowledges that shorter deadlines may put more pressure on maintainers. The company will continue to include guidance with every report and, where possible, a fix or a way to mitigate the risk. Before sending a large number of findings to a single project, it agrees a sustainable pace of work with the maintainers.
Maintainers of critical open-source projects can apply for the free OSS Scanner service. The service regularly scans their code with Anthropic’s most capable models and sends the findings directly to them. Eligible maintainers and project contributors can apply through the Claude for Open Source programme for a free six-month Claude Ultra subscription. Security professionals can apply to the Cyber Verification Program, which adapts Claude’s safeguards for defensive security work.