FIS joins Anthropic Project Glasswing to test Mythos 5 on critical financial software

Fidelity National Information Services, Inc. (NYSE: FIS) has joined Anthropic’s Project Glasswing initiative to test Mythos 5 against software supporting payments, core banking and other critical financial operations. The financial technology company is using the controlled-access frontier model as an additional defensive layer for scanning and evaluating its own systems. The initiative is separate from the companies’ commercial work on agentic banking products, including the Financial Crimes AI Agent scheduled for broader availability during the second half of 2026. The strategic significance lies in FIS applying advanced AI not only to customer-facing workflows but also to the security of the infrastructure on which those services depend. The central tension is whether a highly capable model can materially improve vulnerability discovery while remaining governable, auditable and restricted to defensive use.

How will FIS use Anthropic Mythos 5 to test critical financial-services software?

Project Glasswing is designed for organisations that build or maintain foundational software. Participants receive controlled access to advanced Anthropic models for defensive security work, including the identification and evaluation of potential software vulnerabilities.

FIS plans to use Mythos 5 as an additional layer within its existing security programme. The model will scan and evaluate the company’s own systems rather than being released directly into customer environments as a new banking product.

That scope distinction matters because FIS operates technology that clears payments, moves money and supports core banking for thousands of financial institutions. Security weaknesses in such software can create consequences extending beyond an individual company, including operational disruption, financial loss, data exposure and reduced confidence in interconnected financial infrastructure.

Traditional vulnerability management combines automated scanning, penetration testing, code review, threat intelligence and human security research. A frontier model could potentially expand this process by examining complex code relationships, tracing data flows, reasoning across multiple components and identifying weaknesses that simpler rule-based tools might overlook.

The potential value is not merely faster scanning. Financial platforms frequently contain mature codebases, integrations and customised deployments accumulated over many years. A model capable of reasoning across that complexity could help prioritise vulnerabilities based on their practical relevance rather than producing another large queue of low-value alerts.

However, FIS has not disclosed how much of its software estate Mythos 5 will examine, how findings will be validated or whether the testing has already produced remediated vulnerabilities. Project Glasswing participation should therefore be understood as an active security initiative rather than evidence that FIS has achieved a measurable reduction in cyber risk.

Why is Project Glasswing separate from FIS’s commercial agentic banking programme?

FIS and Anthropic established their commercial agentic AI partnership around the Financial Crimes AI Agent. The product is being co-designed by FIS teams and Anthropic’s Applied AI specialists and forward-deployed engineers.

The agent is intended to automate elements of anti-money-laundering alert and case investigation. It can assemble evidence from banking systems, evaluate activity against known financial-crime patterns and surface higher-risk cases for investigator review. FIS has said the workflow could reduce some investigations from hours or days to minutes while lowering false positives and improving investigative narratives.

BMO and Amalgamated Bank are working with FIS on early deployments, with general availability planned for the second half of 2026. FIS’s longer-term commercial roadmap includes possible agents for credit decisioning, deposit retention, customer onboarding and fraud prevention.

Project Glasswing serves a different purpose. FIS is the user rather than the commercial distributor, and the immediate objective is to strengthen the security of its own technology estate. Mythos 5 is being applied to defensive software analysis, while the commercial programme uses Claude models as reasoning engines within controlled banking workflows.

Separating the programmes provides useful governance clarity. A security-testing model may require access to source code and technical architecture, while a financial-crime agent requires governed access to transaction information, case histories and regulatory procedures. Combining those data environments without strict boundaries could create unnecessary operational and confidentiality risk.

The common strategic principle is controlled deployment. FIS is presenting both initiatives as environments in which data remains governed, actions are traceable and final decisions remain subject to human authority. That approach is especially important in financial services because an AI system that produces an answer without showing the underlying evidence is difficult to use in regulatory, audit or enforcement contexts.

What does controlled frontier AI testing change for banks using FIS platforms?

Banks considering wider AI adoption must evaluate two related questions. The first is whether AI can improve the speed and economics of operations. The second is whether the technology provider can protect the platforms and data required to deliver those improvements.

FIS’s participation in Project Glasswing addresses the second question. Testing its own software with a powerful defensive model may help the company identify weaknesses before they can be exploited or reproduced across multiple client environments.

This is particularly relevant for shared financial technology. A vulnerability in a widely deployed component could affect many institutions, whereas remediation at the platform level could protect a larger portion of the ecosystem. The scale creates operating leverage for both attackers and defenders.

Project Glasswing could also change how FIS incorporates security into software development. Instead of relying primarily on testing near the end of a release cycle, advanced model analysis could be applied during code development, integration and deployment. That would allow vulnerabilities to be investigated before they enter production.

The model should not be treated as an independent security authority. Findings need reproducible evidence, severity assessment and validation by qualified researchers. Remediation must then be tested to ensure that a patch does not create another operational problem.

Banks will also want assurance that customer data is not unnecessarily exposed during model testing. The release says FIS is using Mythos 5 on its own systems, but it does not provide detailed information on data isolation, model retention, deployment architecture or regulatory review.

For financial institutions, the strongest evidence will be practical: fewer critical vulnerabilities reaching production, faster remediation, improved software-release controls and independently verifiable governance. Membership in a controlled initiative is strategically relevant, but it is not a substitute for those outcomes.

Can FIS convert defensive AI capability into a durable competitive advantage?

Cybersecurity is often treated as a cost centre until a major incident demonstrates its commercial importance. For a financial infrastructure provider, however, security capability is closely connected to client retention, contract renewal, regulatory confidence and permission to introduce more autonomous products.

Project Glasswing could support FIS’s competitive position in three ways. First, it can strengthen the company’s internal security processes. Second, it provides experience in governing one of Anthropic’s most capable models. Third, it may reassure banks that the same company offering agentic workflows is also investing in the security of the infrastructure beneath them.

That third benefit is important because agentic AI increases the range of actions software can undertake. A traditional assistant might retrieve information or draft text. An agent can access systems, sequence tasks and initiate actions within defined permissions. The resulting productivity opportunity is larger, but so is the damage that could result from incorrect reasoning, compromised credentials or overly broad access.

FIS already works with the Financial Services Information Sharing and Analysis Center, the Financial Services Sector Coordinating Council, regulators and industry intelligence-sharing networks. Project Glasswing adds another source of technical capability, but the value depends on how effectively findings are integrated into existing controls.

There is also a defensive-versus-offensive tension. A model capable of discovering complex vulnerabilities could be valuable to security teams, but similar capabilities could be misused if access controls fail. Controlled distribution, monitoring and restrictions on model actions are therefore part of the product value rather than peripheral compliance features.

A durable advantage would require more than early access to a particular model. Competing technology providers can also obtain advanced AI capabilities. FIS’s differentiation would need to come from combining model reasoning with its financial infrastructure knowledge, security data, regulatory experience and ability to deploy improvements across a large client base.

How does the Anthropic strategy fit FIS’s post-Worldpay operating and financial reset?

FIS enters this AI expansion following a significant restructuring of its business portfolio. In January 2026, the company completed the sale of its remaining 45% Worldpay interest and acquired Global Payments’ Issuer Solutions business for an enterprise value of US$13.5 billion.

The acquired operation has been rebranded as FIS Total Issuing Solutions and is now part of Banking Solutions. The transaction increased FIS’s exposure to card issuing, account processing and bank technology while removing its remaining ownership connection to the merchant-focused Worldpay business.

The new portfolio makes the Anthropic strategy more commercially relevant. FIS now controls a broader range of systems that connect core banking, issuing, payments, fraud management and regulatory operations. Agentic tools that work across those systems could increase product integration and deepen client relationships.

First-quarter 2026 revenue increased 30% on a reported basis to approximately US$3.3 billion, largely reflecting the acquired issuing operation. Pro forma revenue growth, which provides a more comparable view of the combined business, was 6.5%.

Adjusted earnings increased 12% to US$1.36 per diluted share, while adjusted EBITDA reached US$1.30 billion and the adjusted EBITDA margin rose to 39.6%. Free cash flow more than doubled to US$474 million.

Management maintained its full-year outlook for revenue of US$13.77 billion to US$13.85 billion, adjusted earnings of US$6.22 to US$6.32 per share and free cash flow of US$2.05 billion to US$2.15 billion.

The principal financial constraint is leverage. FIS ended the first quarter with approximately US$21.1 billion of debt and a leverage ratio of 3.6 times. Management is working towards a lower target following the acquisition, which means investment in AI must coexist with integration work, debt reduction, shareholder returns and normal product development.

Project Glasswing itself does not carry disclosed revenue or cost commitments. Its financial significance is indirect. Better security could reduce incident risk and protect customer retention, while experience with Mythos 5 could improve FIS’s ability to build governed commercial AI products.

What does FIS stock performance reveal about investor expectations for the AI strategy?

FIS shares closed at US$42.62 on July 16, rising 3.7% during the session. The announcement was released after the market close, so the gain should not be interpreted as a reaction to Project Glasswing.

The stock was approximately 1.5% above its July 10 close and 7.9% above its June 16 close. That improvement followed a decline to a 52-week low of US$37.42 on June 22.

Despite the rebound, FIS remained 48.4% below its 52-week high of US$82.62. The July 16 closing price gave the company a market capitalisation of approximately US$22 billion.

This performance suggests that the market is assigning limited near-term value to strategic announcements while awaiting clearer evidence from the wider business reset. Investors must assess the integration of Total Issuing Solutions, progress on deleveraging, organic growth and margin delivery alongside emerging AI opportunities.

The depressed share price may also reflect the complexity of comparing reported results following the Worldpay sale and issuing acquisition. First-quarter GAAP earnings included an estimated US$2.2 billion after-tax gain from the Worldpay transaction, making adjusted and pro forma measures more useful for evaluating ongoing performance.

For the AI thesis, the most important financial question is whether FIS can translate technology leadership into additional recurring contract value. Project Glasswing may protect the platform, but the Financial Crimes AI Agent and future banking agents are more likely to determine direct monetisation.

Which milestones will show whether the FIS and Anthropic partnership is working?

The Project Glasswing announcement broadens the FIS and Anthropic relationship from commercial banking automation into defensive software security. It also gives FIS experience with a frontier model whose capabilities require unusually strict access and governance.

What remains unresolved is the operational result. FIS has not disclosed vulnerability findings, remediation times, coverage metrics or a timetable for integrating Mythos 5 into its regular software-development lifecycle.

The next security proof points would be validated vulnerabilities, evidence that material weaknesses were fixed and measurable improvements in detection or remediation. FIS will need to demonstrate that results are reproducible and that the model operates within controlled permissions.

On the commercial side, the Financial Crimes AI Agent remains the nearer revenue catalyst. Broader availability during the second half of 2026, successful deployment at BMO and Amalgamated Bank, and evidence that investigation time falls without weakening accuracy would strengthen the case.

The partnership thesis would improve if FIS can show that security testing and commercial agents share a common, governable architecture while remaining operationally separated. It would weaken if model access proves unstable, findings create excessive false positives, banks delay adoption or FIS cannot show measurable customer economics.

What are the key takeaways from the FIS and Anthropic trusted AI partnership?

• FIS has joined Anthropic’s controlled-access Project Glasswing cybersecurity initiative.
• The company is using Mythos 5 to scan and evaluate its own financial technology systems.
• Project Glasswing is separate from FIS’s commercial deployment of Anthropic-powered banking agents.
• The Financial Crimes AI Agent remains scheduled for broader availability during the second half of 2026.
• BMO and Amalgamated Bank are participating in early commercial agent deployments.
• FIS has not disclosed vulnerability findings, testing coverage or measurable security improvements.
• Controlled access, human validation and auditability are essential because advanced cyber models have dual-use capabilities.
• Project Glasswing could strengthen client trust but does not create disclosed near-term revenue.
• FIS must balance AI investment with Total Issuing Solutions integration and debt reduction.
• Security remediation, bank adoption and recurring contract value are the decisive proof points.