Newshot

AI Finds Vulnerabilities Too Fast: Anthropic Tightens Disclosure Rules, Cutting High-Severity Grace Period From 90 Days to 30

AI summary

Anthropic announced on November 16 that starting January 1, 2027, it will significantly tighten its coordinated vulnerability disclosure policy, shortening the disclosure deadline for "high" and "critical" vulnerabilities from 90 days to 30 days, and to 45 days for other severity levels. The policy applies to vulnerabilities reported to vendors or open-source maintainers on or after January 1, 2027, while reports already in process will follow the original timelines. Anthropic said the previous 90-day buffer was based on the assumption that vulnerabilities could remain private for longer and were unlikely to be independently discovered by multiple researchers, but AI can now find vulnerabilities more efficiently and may also help develop fixes, raising the risk of duplicate discovery. In addition to shortening the disclosure window, Anthropic will speed up its response process: if maintainers do not respond within 15 days of the initial report (previously 30 days), the company will seek outside vulnerability coordination bodies earlier; after a patch is released, Anthropic will wait 30 days (previously 45 days) before publishing full technical details; and the 7-day disclosure dead

Translation(model translation, not the source's own text)
  • Day/Night Follow system Light Dark

AI Finds Vulnerabilities Too Fast: Anthropic Tightens Disclosure Rules, High-Risk Vulnerability Buffer Period Changes from 90 Days to 30 Days

IT之家, November 17: Anthropic yesterday (November 16) issued an announcement announcing that starting January 1, 2027, it will significantly tighten its "Coordinated Vulnerability Disclosure" policy, adjusting the disclosure period for "high" and "critical" vulnerabilities from 90 days to 30 days, while narrowing other levels of vulnerabilities to 45 days.

Citing a blog post, IT之家 reports that the policy applies to vulnerabilities reported to vendors or open-source maintainers on or after January 1, 2027; reports already in the processing pipeline will still follow the original timelines.

Anthropic said the previous 90-day buffer period was based on the assumption that "vulnerabilities can remain private for a relatively long time," and that the probability of the same vulnerability being independently discovered by different researchers was low. But today, AI can not only discover vulnerabilities more efficiently but may also help develop fixes, and the risk of vulnerabilities being discovered repeatedly rises accordingly.

In addition to shortening the disclosure window, Anthropic will also speed up its response process. If maintainers do not respond within 15 days (previously 30 days) after receiving the initial vulnerability report, the company will seek intervention from an external vulnerability coordination organization earlier. At the same time, after a patch is released, Anthropic will wait 30 days (previously 45 days) before making full technical details public. For vulnerabilities being actively exploited, the 7-day disclosure deadline remains unchanged.

The new policy also takes into account the reality that some systems are difficult to fix quickly. Vulnerabilities involving critical infrastructure or requiring changes to technical specifications or RFCs may not be fixable within 30 or 45 days; Anthropic said it will negotiate a reasonable remediation timetable with the relevant vendors.

Anthropic also announced that for high-risk and critical vulnerabilities that have exceeded the original 90-day deadline but have not yet been disclosed, maintainers will be given an additional 30-day remediation period; starting January 1, 2027, unfixed reports will be disclosed on its vulnerability disclosure platform under the new policy.

  • "Responding to the Challenge of AI Vulnerability Discovery: Google Project Zero Plans to Gradually Shorten Vulnerability Public Disclosure Deadline to 30 Days"

Related Articles

  • Claude API adds ability to adjust effort: lets AI apply effort according to the task, can "shift gears" midway from casual chat to complex analysis
  • Anthropic upgrades AI office suite: from brainstorming to data dashboards, Claude wants to take over the entire team collaboration workflow
  • "Programming Interviews Are Dead": Anthropic Says Coding Questions Can No Longer Screen Programmers in the AI Era
  • Anthropic unveils unreleased AI model: more capable than Mythos 5.1, and more "well-behaved"
  • Anthropic report: AI work hours are already 6 times those of humans, junior programmer positions plummet, revenue per employee is 2 to 4 times that of Meta / Google in the same period
  • Anthropic test incident: Claude model submitted false murder leads to police and also submitted 20 visa applications

Sites under 软媒: IT之家 最会买 - Rebate and Cashback Coupons iPhone之家 Win7之家 Win10之家 Win11之家

Software under 软媒: 软媒手机APP应用 魔方 最会买 要知

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗