Anthropic Accelerates Vulnerability Disclosure Timelines in Response to AI-Driven Threat Landscape

In a significant shift that reflects the rapidly accelerating pace of cybersecurity threats, AI research company Anthropic has announced a major revision to its vulnerability disclosure policy. Starting January 1, 2027, the company will dramatically shorten the window of time it grants software vendors and maintainers to address serious security flaws before it publicly discloses the findings. The update cuts the standard remediation period for critical vulnerabilities from 90 days down to just 30 days, a move that signals the company’s acknowledgement that the traditional "human-speed" disclosure model is no longer sufficient to combat the speed at which modern AI models can identify and exploit software weaknesses.

The change, which was detailed in a November 16 post by Anthropic’s Frontier Red Team, applies specifically to vulnerabilities identified by the company’s internal AI models and subsequently reported to vendors and project maintainers. It does not apply to external security researchers who report bugs to Anthropic, nor does it affect reports currently in progress, which will continue to be governed by the legacy timelines until their resolution.

The Death of the "Human-Speed" Assumption

For years, the cybersecurity industry has operated under the assumption that vulnerability disclosure should allow for a generous grace period to ensure that developers have sufficient time to craft, test, and deploy patches. This model was predicated on the belief that most software vulnerabilities were complex, difficult to discover, and required significant human effort to replicate. Anthropic’s leadership, however, has concluded that these foundational assumptions have been rendered obsolete by the rise of powerful, automated AI systems.

"Disclosure norms, including our own, had assumed fixes moved at human speed and that independent rediscovery was unlikely," the Frontier Red Team noted in its announcement. "Neither assumption holds today."

The company highlighted the growing reality of AI-driven vulnerability research, noting that a security flaw discovered by one model is increasingly likely to be detected by another, often in a fraction of the time it would take a human researcher. As these models become more adept at scanning code, identifying edge cases, and mapping potential exploit paths, the risk that a malicious actor might independently discover and weaponize a vulnerability before a fix is released has grown exponentially. By shortening the disclosure window, Anthropic is attempting to force a corresponding acceleration in the patch management lifecycle, ensuring that developers prioritize the security of their software in an environment where speed is a critical defense mechanism.

Structural Changes to the Disclosure Policy

The revised policy introduces a tiered structure for disclosure, differentiating between the severity of the findings. Under the new guidelines, high- and critical-severity vulnerabilities are now subject to a 30-day disclosure deadline. For all other issues, which are deemed to be of lower severity, the window is set at 45 days.

Recognizing that the initial communication phase is often a bottleneck, Anthropic has also tightened its escalation protocols. The company will now aim to escalate unresponsive reports to an external vulnerability coordinator after just 15 days of silence from a maintainer—a significant reduction from the previous 30-day threshold. This ensures that if a maintainer is failing to engage with a security report, third-party oversight can be brought in much earlier to facilitate a resolution.

Furthermore, the company is refining how it handles the public release of technical details. Previously, Anthropic would wait 45 days after a patch was made available before publishing the full technical breakdown of the vulnerability. That period has been shortened to 30 days. This adjustment is designed to balance the need for public awareness with the need for developers to have a reasonable amount of time to push updates to their user base. However, the policy remains consistent in one key area: for vulnerabilities that are already under active exploitation in the wild, the seven-day disclosure window remains unchanged, reflecting the urgency required when a threat is already causing harm.

Nuance in Implementation and Scope

While the new policy represents a more aggressive stance on security, Anthropic has built in specific provisions to account for the realities of complex software environments. The company explicitly stated that it will work collaboratively with vendors to establish bespoke remediation timelines for "hard-to-deploy" software. This includes critical infrastructure, where the deployment of a patch can carry significant operational risk, or scenarios where a vulnerability requires fundamental changes to specifications or RFCs (Request for Comments).

Moreover, the policy contains a specific carve-out regarding the source of the vulnerability report. Anthropic noted that the new, accelerated timelines do not apply to findings generated by its OSS Scanner that have not undergone thorough human review. This distinction serves as a safeguard against the potential for "noise" or false positives generated by automated tools to force unnecessary, rushed work on maintainers.

Acknowledging that these shorter timelines place a significant burden on the maintainers responsible for fixing these bugs, Anthropic has committed to providing more than just a list of issues. The company intends to continue sending actionable guidance alongside each report, and, wherever possible, will provide a suggested patch or mitigation strategy. By doing the "heavy lifting" of the remediation research, Anthropic hopes to lower the barrier for maintainers to fix the issues within the truncated 30-day window.

Industry Comparison and the Evolving Landscape

Anthropic’s decision to move toward a 30-day disclosure cycle puts it at the vanguard of a broader industry shift, placing it ahead of established frameworks such as Google’s Project Zero. Project Zero, which has long been a standard-bearer for disclosure transparency, currently publishes bug details 90 days after a report is filed, regardless of whether a patch has been released.

However, even the industry titans are moving toward shorter timelines. Google has announced plans to cut its own 90-day window to 60 days by February 2, 2027, and eventually to 30 days by January 3, 2028. Anthropic’s policy effectively accelerates this transition, aligning its disclosure posture with a future where the gap between vulnerability discovery and exploitation is closing rapidly.

There remains a philosophical disagreement regarding the role of severity in these timelines. While Anthropic has opted for a tiered approach, setting different windows based on the perceived impact of the vulnerability, Project Zero has traditionally argued that a vulnerability’s severity is largely irrelevant to the likelihood of it being found by an attacker. From their perspective, once a bug is discovered, the race between the defender and the attacker begins, and the severity does not necessarily influence the speed at which an attacker can build an exploit.

The urgency driving these policy changes is underscored by external developments in the threat landscape. On the same day Anthropic announced its policy update, Google’s Threat Intelligence Group reported the discovery of an exposed server hosting an "exploit_agent" framework. This tool was specifically designed to automate the research of vulnerabilities and the subsequent weaponization of those flaws. The existence of such tools confirms the fears held by many in the security community: the automation of the attack chain is no longer theoretical; it is a live and evolving threat.

In a final effort to manage the transition to this new policy, Anthropic has stated that it will grant maintainers an additional 30 days on any high- or critical-severity bugs that have already passed its current 90-day deadline. This grace period serves as a bridge, ensuring that the shift does not immediately force a flood of disclosures onto a dashboard that is designed to keep the security community informed about the state of software vulnerabilities. By balancing its need for speed with a pragmatic approach to the human element of software development, Anthropic is attempting to set a new standard for a world where AI discovery is the new normal.