Redefining Financial Cybersecurity
Three concurrent announcements between April and May 2026 reconfigure the economics of financial cybersecurity. Dominican banking has a window of twenty-four months.

Between April and May 2026, three concurrent announcements reshaped the economics of cybersecurity and the technology architecture of banking and the global financial sector.
First, Anthropic presented Claude Mythos, a frontier model that found a twenty-seven-year-old flaw in OpenBSD for less than twenty thousand dollars in total compute; the specific run that discovered it cost less than fifty. Second, OpenAI opened GPT-5.5-Cyber to vetted defenders through a scheme called Trusted Access for Cyber. And finally, EPAM and Anthropic signed a multi-year alliance with an explicit focus on regulated banking and the modernization of legacy cores. Three consecutive moves that mark the start of the twenty-four-month cycle in which Dominican banking will decide its competitive position for the next decade.
The main axis is the collapse in the marginal cost of discovering vulnerabilities. A serious pentest for a mid-sized Dominican bank —web application, APIs, cloud infrastructure, internal network— costs today between twenty-five thousand and eighty thousand dollars, with timelines of thirty-five to one hundred days from contracting to delivery of the report.
The international benchmark 2026 published by Blaze Information Security and TCM Security puts the ranges in that same order. XBOW, the company that in June 2025 took first place on HackerOne's U.S. leaderboard with one thousand and sixty vulnerabilities reported in ninety days, launched in November Pentest On-Demand: the equivalent of two to four weeks of manual work delivered in a few days, at a significantly lower price according to its own commercial communications.
In internal benchmarks against five pentesters professionals —the most senior with more than twenty years of experience— XBOW solved in twenty-eight minutes what the human team solved in forty hours. The Series C of one hundred fifty-five million dollars closed in May, with Accenture Ventures, NVIDIA's NVentures, Samsung Ventures and SentinelOne's S Ventures, validates its commercial trajectory.
Mythos operates on another plane. It found a seventeen-year-old flaw in FreeBSD's NFS that allows unauthenticated root access from the open internet. It identified a sixteen-year-old flaw in FFmpeg's H.264 decoder, a library fuzzed by the industry for a decade. It built chains of four vulnerabilities to escape browser sandboxes, a class of attack that takes senior researchers months. More than ninety-nine percent of the thousands of bugs found remained unpatched at the time of the announcement.
The UK AI Security Institute measured a seventy-three percent success rate in capture-the-flag challenges at expert level. Anthropic sells Mythos at five times the price of Opus 4.6, restricted to partners in the Glasswing program —Amazon, Apple, Cisco, Google, JPMorgan Chase, Microsoft, NVIDIA, Palo Alto Networks. The window in which that capability remains exclusively in defensive hands is estimated at between twelve and eighteen months, according to Anthropic itself, Wiz and Oege de Moor, founder of XBOW. The time-to-exploit median, according to ArmorCode, fell from seven hundred seventy-one days in 2018 to less than four hours in 2024 and is projected to drop below one hour by the end of 2026.
Dominican banking enters this cycle with figures of its own that deserve attention. The Superintendencia de Bancos reported forty thousand eight hundred operational fraud events in 2025, with gross losses of 2,498 million pesos. Gross losses have grown by 55.6 percent since 2021. E-commerce displaced internet banking as the channel with the largest materialization, with seven hundred twenty-two million pesos in losses in 2024.
On the other hand, FortiGuard recorded five hundred thirty-five million attempted cyberattacks against the country in 2025. Sixty percent of financial institutions increased their cybersecurity investment compared to the previous year, according to a statement by Superintendent Alejandro Fernández. Adoption of artificial intelligence in banking reached fifty-six percent of institutions in 2025, up from twenty-seven percent in 2021. The country fell from seventh to ninth place in the Latin American Artificial Intelligence Index between 2024 and 2025, overtaken by Costa Rica and Colombia.
The second axis is enterprise deployment architecture. On 6 May, EPAM and Anthropic announced a multi-year partnership covering Anthropic's full technology stack —Claude, Claude Code, Claude Agent SDK and Claude Security— and a dedicated practice of more than ten thousand certified architects, including two hundred fifty forward-deployed engineers known as Black Belts.
In this sense, EPAM completes an Anthropic trifecta in which Deloitte contributes four hundred seventy thousand users in one hundred fifty countries and Accenture contributes thirty thousand trained professionals with its own unit, the Accenture Anthropic Business Group. EPAM's differentiation is engineering: Eastern European heritage, strength in capital markets, nine years serving UBS. Its entry into the Anthropic stack is optimized for regulated banking with compliance concerns and legacy core modernization. The trifecta consolidates a de facto standard for the next twenty-four months: AWS plus Anthropic plus one of Accenture, Deloitte or EPAM as the implementation layer.
That de facto standard raises a concrete operational question for Dominican banking. The five large multiple-service banks will have direct coverage from the tier-one integrators. Mid-sized banks and the savings and loan associations will depend on a local market of integrators certified in Claude, Anthropic or equivalents. Whoever captures certification locally captures the decade. InformationWeek documented in December that the triangle of model provider plus cloud plus integrator creates architectural pressure: the client ends up tied to a specific stack for five to seven years. The decision on which integrator to choose in 2026 and 2027 fixes technological dependencies until 2032.
From the standpoint of Tabuga, corporate executives face three possible moves in the next quarter. The first is to diagnose Mythos exposure: audit the core banking system to identify software dependencies with a high probability of harboring vulnerabilities of the OpenBSD-FreeBSD-FFmpeg class, where flaws of fifteen to twenty-seven years survive in code reviewed for decades. The second is to design machine-scale patching pipelines that combine static and dynamic analysis, autonomous agents contracted as a service, human governance for high-impact cases, and structured reporting to the Superintendency. The third is to build the local governance layer that no global integrator can offer at equal depth: simultaneous compliance with the Monetary Board's Cybersecurity Regulation, Ley 53-07, the National Securities Market Council Regulation R-CNMV-2024-08-MV, circular CSB-RG-202500015 and the new Operational Risk Regulation published in April.
Decisions made this quarter commit the operating balance of Dominican financial institutions until 2032. A bank that signs a contract with an autonomous pentesting provider before the first quarter of 2027 will enter the new regime with continuous coverage. A financial institution that keeps the annual cycle of pentests manual as formal regulatory compliance will bear a growing opportunity cost, measured in zero-days undiscovered and in accumulated operational losses.
«We would like to help companies protect themselves, and we think it is important to start working on this quickly», said Sam Altman, CEO of OpenAI, addressing the issue on X last month.
Through the technical network of Tabuga, we see as necessary the intermediate layer where the technical depth of the frontier model converges with local regulatory knowledge and the continuous auditing demanded by regulators. The calculation is as straightforward as the figures allow.