Project Glasswing partners can now share Mythos findings beyond the programme

At a glance:

• Anthropic revises Project Glasswing policy to allow partners to share Mythos cybersecurity findings with external parties
• Mythos has identified thousands of zero-day vulnerabilities across major operating systems and browsers
• The change comes as Anthropic navigates regulatory scrutiny from financial institutions and government agencies

Policy Change for Project Glasswing

Anthropic announced on Monday that it is revising its disclosure policy for Mythos, the unreleased cybersecurity-focused AI model deployed under its Project Glasswing controlled-access programme. The revision will allow partners using Mythos to share information about cyber threats with other potentially exposed parties, rather than keeping findings within the original partner organization. This represents a significant shift from the company's previous approach, which had been substantially tighter with findings held inside the partner programme and surfaced only to Anthropic itself.

The list of parties partners can now share with is deliberately broad, according to the published text. This includes security teams at other companies, industry bodies, regulators and government agencies, open-source maintainers, the media and the public. All such sharing remains subject to responsible-disclosure norms, which typically include reasonable patching windows and constraints on weaponizable details. The defender pool has effectively expanded beyond the original partner organizations to create a more comprehensive security ecosystem for addressing emerging threats.

Mythos Vulnerability Discoveries

The policy change carries particular significance due to what Mythos has been uncovering during its deployment. According to Anthropic's own disclosures, the model has identified thousands of zero-day vulnerabilities across major operating systems and browsers in internal testing. More impressively, it has demonstrated the ability to develop working exploits against those flaws on first attempt in more than 83% of cases. These capabilities position Mythos as a potentially transformative tool in the ongoing battle against cyber threats.

Project Glasswing's partner roster includes some of the most prominent technology and financial institutions in the world: Amazon Web Services, Apple, Google, Microsoft, Nvidia, Cisco and JPMorgan. This group is large enough that the findings circulating within it represent a meaningful subset of the modern enterprise attack surface. The collective knowledge base now accessible through the expanded sharing policy could significantly enhance security across industries by enabling more rapid identification and remediation of vulnerabilities.

Regulatory Context

The disclosure-policy loosening arrives amid a wider regulatory arc that Anthropic has been navigating. The company is preparing to brief the Financial Stability Board on what Mythos has discovered within financial-services infrastructure, at the request of Bank of England Governor Andrew Bailey. This briefing reflects the growing importance of cybersecurity in financial stability discussions and the recognition that advanced AI tools like Mythos could play a crucial role in identifying systemic risks.

The regulatory cohort extends beyond the Financial Stability Board to include ASIC (Australian Securities and Investments Commission), the Federal Reserve, the Bank of England, the European Central Bank, the US Treasury and several Asian regulators. These institutions are part of a coordinated monitoring effort that recognizes the potential impact of undiscovered vulnerabilities in financial systems. The policy change appears consistent with what regulators in these forums have been privately advocating for: that vulnerability findings should not be held tightly within a partner programme that excludes most of the financial supervision community.

Pentagon Deployment and Operational Context

A parallel operational track worth highlighting involves the Pentagon's deployment of Mythos. According to the Defense Department's top technology official, the military has been using the AI model to find and patch software vulnerabilities across the US government. This deployment represents a significant vote of confidence in Mythos's capabilities, though it comes alongside a complex Trump administration posture toward Anthropic itself. The Pentagon is reportedly racing to complete a transition away from Anthropic even as it continues to leverage the company's technology.

In another development, UK banks received their own Mythos briefing earlier this month. The new partner-sharing rules enable those briefings to flow downstream in ways that the previous structure prevented. This suggests that while access remains controlled, the flow of information is becoming more dynamic and responsive to the needs of different sectors and organizations. The expanded sharing framework could prove particularly valuable in highly regulated environments like finance, where coordinated responses to emerging threats are essential.

Responsible Disclosure Mechanics

The technical and policy mechanics of responsible disclosure remain the limiting factor in how broadly Mythos findings can be shared. Anthropic's revised policy does not abandon the requirement that downstream sharing be coordinated under standard responsible-disclosure norms. These norms typically include reasonable timeframes for vendors to develop patches before detailed vulnerability information becomes public, as well as constraints on sharing information that could be weaponized by malicious actors.

The emphasis on responsible disclosure reflects a balance between the urgent need to address security vulnerabilities and the potential risks associated with prematurely revealing information that could be exploited. This approach acknowledges that while transparency is valuable in the security community, it must be balanced against the potential for harm if vulnerability information falls into the wrong hands before appropriate mitigations are in place.

Structural Asymmetry Concerns

What the change does not resolve is the structural asymmetry that critics have been highlighting for two months. The 40-to-50 organizations within Project Glasswing continue to receive the defender's view of Mythos before the rest of the global enterprise ecosystem. Rather than collapsing through wider direct access, this asymmetry is now expanding through the new partner-sharing rules. This means that while information is flowing more freely among partners, the broader community still lacks direct access to the model's capabilities.

Anthropic has consistently framed its controlled-access programme as calibrated to give defenders a head start on the inevitable adversary use of comparable capabilities. The loosened sharing rules represent the most concrete operational evolution of this framing since Mythos was first announced in April. However, critics argue that the fundamental tension between controlled access and widespread benefit remains unresolved, and that the current approach may create an uneven playing field in the cybersecurity landscape.