Bangladeshi financial sector’s cyber resilience in the age of autonomous AI

For nearly two decades, cyberattacks against financial institutions required skilled human operators bound by human speed, working hours, and the limited number of qualified attackers in the world. The capabilities AI laboratories are now disclosing suggest this constraint is loosening.

In April 2026, Anthropic launched Claude Mythos Preview, a model the company describes as capable of identifying and helping fix software vulnerabilities at a scale no human team could match. In early testing, Mythos identified a security flaw in OpenBSD that had gone undiscovered for twenty-seven years. More unsettlingly, during internal safety testing, an earlier version of the model escaped its sandbox, gained unsanctioned internet access, and emailed the supervising researcher to report its own success — an action neither requested nor expected. 

Anthropic disclosed Mythos alongside Project Glasswing, a defensive initiative now spanning roughly 150 organisations across more than fifteen countries, including AWS, Apple, Cisco, JPMorgan Chase, Microsoft, and Palo Alto Networks. On 9 June, the company released a public, hard-limited version under the name Claude Fable 5.

The frontier keeps moving. In May 2026, Microsoft disclosed MDASH, a multi-agent system that topped the CyberGym benchmark and found sixteen unknown Windows vulnerabilities. OpenAI has introduced a comparable programme, Daybreak. The working assumption for any regulator: tools that autonomously find and exploit software weaknesses will keep growing more capable and more widely available — and will eventually reach actors with harmful intent.

Bangladesh has made real progress here. Bangladesh Bank's ICT Security Guideline set baseline expectations on audit, patching, and incident response, and its Cybersecurity Framework, Version 1.0 (2026), now creates a stronger mandatory baseline for banks, NBFIs, and payment operators, with implementation due by end-2026. The question is no longer whether Bangladesh has rules, but whether they can be recalibrated fast enough for an AI-shaped threat environment.

A centralised registry of core banking system versions and patch status, housed under Bangladesh Bank's new Risk-Based Supervision framework, would give supervisors visibility they currently lack. BSEC's regulatory perimeter — covering the exchanges, the Central Depository, and broker-dealers — needs comparable cybersecurity attention as automated trading and mobile-based access expand the attack surface.

Three structural realities make this urgent. First, the 2016 theft of $81 million from Bangladesh Bank's Federal Reserve account, executed via fraudulent SWIFT instructions after months of human-directed planning, showed how a single point of compromise can carry national consequences — automated tools now make that kind of reconnaissance far faster. 

Second, Bangladesh's core banking systems are uneven in age and configuration, with many institutions running legacy environments layered with newer modules nobody fully maps any more — exactly the terrain automated scanning tools exploit best. Third, the financial system extends well beyond scheduled banks, into MFS providers, microfinance institutions, and capital market intermediaries; the system's security is only as strong as its least-resourced participant.

Several jurisdictions offer useful models. The EU's Digital Operational Resilience Act mandates intelligence-led penetration testing and curbs over-reliance on single vendors. Singapore's tiered technology risk guidelines scale expectations by institutional size. The Bank of England's CBEST pioneered threat-intelligence-led red-teaming, while the US has built coordination mechanisms through CISA and a risk framework through NIST. The common thread is institutional, not technological: coordination, information-sharing, and proportionate rules that adapt as threats evolve.

Bangladesh's foundations exist; the task is evolving them. Supervisory frameworks at Bangladesh Bank and BSEC need to translate AI-augmented threats into measurable obligations — patch timelines matter far more when adversaries operate at machine speed. A formal, real-time threat-sharing mechanism, modelled on Singapore's FS-ISAC participation, with legal protection for institutions sharing in good faith, would cut the lag between a threat surfacing at one institution and action at others. 

A centralised registry of core banking system versions and patch status, housed under Bangladesh Bank's new Risk-Based Supervision framework, would give supervisors visibility they currently lack. BSEC's regulatory perimeter — covering the exchanges, the Central Depository, and broker-dealers — needs comparable cybersecurity attention as automated trading and mobile-based access expand the attack surface. 

Supply-chain risk also warrants structured assessment rather than blanket exclusions, given how much of the sector's hardware and software depends on foreign vendors. And the data-localisation requirement already written into the Personal Data Protection Act 2026 needs financial-sector translation: which systems count as critical information infrastructure, and what audit standards apply.

This is not a case for panic. Existing controls are not worthless; the bar they need to clear is simply higher, and the timeframe shorter. A sensible sequence would start with an inter-agency briefing between Bangladesh Bank, BSEC, the Bangladesh e-Government Computer Incident Response Team, and the finance ministry, followed by a supervisory note on AI-era threats, a threat-sharing mechanism, a CBS version registry, and a structured vendor-risk process — each achievable under existing legal authority.

The 2016 SWIFT heist cost Bangladesh $81 million in stolen funds and years of reputational repair. A measured, multi-year programme to strengthen the sector's cyber posture costs far less by comparison. Waiting for the next incident to set the agenda is not a strategy — and right now, the choice between acting early and reacting late still belongs to Bangladesh.

Rokibul Islam Bin Yousuf, FMVA, CERM, is an investment banking professional at Prime Bank Investment PLC

Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the opinions and views of The Business Standard.