Issue 12
Executive Signals | 18 November 2026
Whenever a financial institution reports a cybersecurity incident, I keep coming across the same phrase. The core banking system was not affected. We are often told, too, that customer accounts remain safe, business operations are continuing and the incident has been contained.
I have a problem with what we assume when we hear this. If the core banking system is safe, is the institution secure? Does the safety of the core application establish the security of the financial services that depend on it?
Two incidents from last month make me question these assumptions.
October 2026 · two Indian banks
187 unauthorised transfers from its settlement account with the RBI, between 15 and 17 October. Cause not established publicly.
Customer records with personally identifiable information, found in about 3.75 lakh records circulating on the dark web. Disclosed on 21 October.
As reported by the banks and in the police complaint. Both investigations are continuing.
Between 15 and 17 October, Central Bank of India had an incident in its outward RTGS system. On 19 October the bank described it as a technical glitch, confirmed that the affected systems had been isolated and said customers had suffered no financial loss.
A police complaint later obtained by the media described about 187 unauthorised transfers totalling ₹13.97 crore from the bank’s settlement account with the Reserve Bank of India. It alleged unauthorised access involving the Structured Financial Messaging System (SFMS), which supports interbank financial messaging.
I found it odd that the entries in the bank’s normal transaction systems did not match the complaint. Some partial logs showed up in the SFMS logs. The bank later said it had identified a problem with its local RTGS interface and restored services. The precise technical cause of the unauthorised transfers has not been established publicly, so it would be wrong to conclude that the central SFMS infrastructure was compromised.
On 21 October, State Bank of India disclosed an incident involving the exposure of customer information. Its preliminary examination of about 3.75 lakh records circulating on the dark web found 3,897 records containing personally identifiable information. SBI engaged independent investigators and said its core business operations and continuity were not materially affected.
The incidents were different. One involved disputed financial transactions and the other customer information. But they point to a problem that has been developing for years. Attackers have not stopped attacking core systems. They have found that they often do not need to.
A financial institution can protect its core applications and still be exposed through the infrastructure, software, identities and services around them. Attackers want money, information and business processes, and they do not have to follow the routes we have spent years defending.
The Core Is Not the Boundary
Financial institutions have invested heavily in securing their infrastructure. Core applications get special attention, and for good reason. They handle transactions, maintain records, support accounting and process activities the institution cannot do without.
Over the years we have built layers of controls around these systems. Network segmentation, privileged access management, vulnerability management, endpoint security, monitoring, penetration testing and audits have made direct compromise harder. Many banks run security operations that can identify and respond to attacks that would have gone unnoticed a decade ago.
The financial system has changed alongside those improvements. A bank’s operations now depend on payment processors, financial messaging systems, identity providers, cloud platforms, software libraries, authentication services, customer-facing applications and third-party technology providers. Each dependency brings its own infrastructure, permissions and operational relationships.
Some of these dependencies are visible and get scrutiny. Others are buried inside applications, software components and service integrations. An institution might know it uses a particular payment gateway and still know little about the components that gateway depends on. It might understand the security of its own web application and have less visibility into the JavaScript a third-party service delivers to customers’ browsers.
We have traditionally treated many of these matters as application security, vendor management or infrastructure maintenance. I think that separation is becoming a problem. An attacker does not distinguish between a core application and a supporting service the way we do. If compromising an identity provider gives access to customer information, why attack the customer database? If a trusted integration can initiate transactions, why spend months trying to penetrate the core banking system? The target is the outcome, and the route is whatever gets the attacker there.
There is another assumption I see often. Most organisations have transferred a large part of their technology risk to outside vendors, and they hope those vendors will be accountable when something goes wrong. They will not be. When an incident reaches customers, regulators and the public, everyone looks at the bank or the financial institution, no matter who it outsourced the work to.
Work can be outsourced, responsibility cannot.
The Infrastructure We All Depend On
The vulnerabilities disclosed in PHP last month are another example. PHP is one of the languages on which much of the web was built. It runs behind websites, business applications, content management systems and enterprise products, and organisations using these applications may not know PHP is part of their technology stack.
On 21 October 2026, the PHP development team released emergency security updates for two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619, together called PHP2Shell. The flaws could allow unauthenticated remote code execution in vulnerable installations. The concern went beyond websites running business applications, because PHP is also embedded in products that organisations use to manage and protect their infrastructure.
On 12 November, Check Point Research published details of attacks that began on 30 October against vulnerable Security Gateways. Attackers exploited PHP2Shell through internet-facing Mobile Access portals, chained it with a privilege-escalation vulnerability and established persistent access. In one investigated intrusion, they moved from the gateway to a VMware ESXi host and deployed destructive malware.
Consider the route. The attacker did not begin with the application’s business logic or database. The entry point was a security gateway, a component organisations deploy to protect access to their environments.
Security advisories identified affected products from Palo Alto Networks, Cisco, Citrix, F5 and Fortinet, although exposure varied by product and version. Confirmed exploitation was reported against selected Palo Alto Networks and Check Point gateways. An affected product is not the same as a confirmed compromise. Still, a vulnerability in a programming language used by a wide range of products can create exposure in places security teams may not associate with that language. We depend on layers of infrastructure that are common across organisations and industries, and when a weakness appears in one of them, the number of systems needing attention can be enormous.
PHP is only one example. Earlier this year, Microsoft reported a supply-chain campaign involving more than 400 packages in the JavaScript npm ecosystem. The attackers distributed malicious package versions that could execute code during installation, steal developer and cloud credentials, and spread through compromised publishing accounts. Developers and automated build systems use these packages to create other applications, so an organisation could have secure production infrastructure and still expose credentials when a developer or build pipeline installed a compromised dependency.
These are two different problems. PHP2Shell was about exploitable vulnerabilities in software used by applications and security products. The npm campaign was about malicious code entering through the software distribution process. Neither establishes that a particular bank’s core systems were compromised, and that is not my argument. They show how an attacker can begin with an overlooked component and end up with access to systems and permissions that matter to the business.
A Trusted JavaScript File and 178 Organisations
On 16 November, another incident showed how far a compromise of shared infrastructure can reach. Scott Helme of Report URI published an investigation into malicious JavaScript being delivered through Optimizely, a platform organisations use to experiment with and personalise their websites.
Optimizely’s Web Experimentation product works by delivering JavaScript to customer websites. This lets organisations test changes, compare user experiences and modify how pages behave. For the businesses using it, that is a legitimate part of running a website.
According to Optimizely’s own incident report, an attacker exploited a vulnerability in its account signup process. The flaw allowed the attacker to impersonate users and insert malicious code into customer projects without stealing their passwords. The code was then served through Optimizely’s legitimate infrastructure, from a trusted domain, through the expected delivery mechanism and with a valid security certificate.
Report URI analysis of Optimizely snippets
The full extent of visitor exposure and financial losses is not known.
Report URI examined 1,611 Optimizely JavaScript snippets and found 188 infected files belonging to 178 organisations. Report URI’s list includes web properties of major technology, media, retail and financial-services brands, among them HSBC and Equifax. That means affected Optimizely snippets were associated with their websites. It does not mean their own networks were breached or that their customer data was stolen. The malicious code was designed to replace cryptocurrency wallet addresses, capture payment-card information from accessible forms, collect certain browser information and load additional exploits targeting vulnerable iPhones and iPads. Researchers identified cryptocurrency theft linked to the campaign, although the full extent of visitor exposure and financial losses remains unknown.
Optimizely confirmed that the problem affected its Web Experimentation product, said it had closed the vulnerability and removed the injected code, and stated that the exposure was through visitors’ browsers and not customers’ networks.
Think about what that means. A business could have kept a secure website, patched its servers, protected its databases and followed its normal security procedures, and its visitors could still receive malicious code because a trusted service was delivering it. A bank using such a service would not need to have its core banking platform compromised for its customers to meet malicious code on its website. The attacker would be exploiting the trust the website placed in a third-party script.
A valid certificate establishes the identity of the server delivering the code. It does not establish that the code is safe.
That distinction is familiar to security professionals, but I wonder how often it features in the assurances given to management.
There was evidence that preventive controls mattered. Report URI found that restrictive Content Security Policies blocked the malicious payload on some affected sites, while others allowed it to execute. Some policies stopped data from being sent to the attacker’s servers even where the code ran. Defenders did have options here. The security of a website depended partly on controls governing code delivered by another organisation.
An A/B testing and personalisation tool turning into a route for payment-card theft is not something most executives expect to see in their cybersecurity risk discussions. Perhaps it should be.
The Attackers Follow the Permissions
The attacks on Binance and Crédit Mutuel Arkéa make the argument from another direction. On 29 October, cryptocurrency exchange Binance was hit by a theft that affected about $6.34 billion in digital assets, and Binance says about $2.85 billion had been frozen or intercepted by 3 November. The FBI attributed the attack to North Korea’s TraderTraitor operation.
Binance’s preliminary investigation described a campaign prepared over several months. The attackers used social engineering, malicious software disguised as development tools, compromised employee devices and stolen credentials to get into internal systems. One engineer had downloaded a tampered software installer while trying to install an AI coding tool. Another was targeted through a fraudulent recruitment process that included a malicious coding assignment. The attackers then used compromised devices and an internal AI tool to reach more engineers and obtain deployment permissions.
The route into Binance, as described in its preliminary investigation
Access at each layer created authority at the next.
Those permissions were used to deploy a malicious container and reach wallet-signing operations. Binance reported that the attackers used its wallet-signing service to establish malicious EIP-7702 authorisations without obtaining the wallet’s private key. Those authorisations continued to operate independently of its normal withdrawal infrastructure, which meant suspending withdrawals was not sufficient to contain the attack.
I find the sequence instructive because access obtained at one layer created authority at another. The attackers began with people and tools, moved through development and deployment permissions, and reached the systems that can authorise financial transactions.
Six days later, on 4 November, French banking group Crédit Mutuel Arkéa disclosed a different attack. Attackers used the platform of an external service provider to reach the bank and execute fraudulent transfers from the bank’s own account. The bank said the attackers had not come in through its online banking subsidiary, Fortuneo, and that customer accounts and assets were unaffected. Its security systems detected the activity and it took steps to contain the incident.
These are different financial businesses running different technologies and facing different attacks, so I would not turn them into examples of one technical vulnerability. What they share is that the attackers used something the businesses already trusted. At Binance it was employee devices, tools and deployment permissions. At Crédit Mutuel Arkéa it was a service-provider platform. Understanding the permissions that come with a business relationship matters as much as understanding the vulnerabilities in the systems themselves.
What the Numbers Tell Us, and What They Don’t
Research published during 2026 supports the concern about third-party exposure. The Verizon Data Breach Investigations Report examined 3,809 incidents in the financial and insurance sector, 1,300 of them with confirmed data disclosure. Third-party involvement featured in 34% of the sector’s breaches in its dataset. Visa’s Fall 2026 Biannual Threats Report, released on 17 November, reported a 62% increase in incidents involving third-party agents compared with the preceding six months. These providers were associated with about 56% of the compromised payment credentials identified during the period.
Third-party exposure in two separate datasets
Verizon Data Breach Investigations Report. 3,809 incidents, 1,300 with confirmed data disclosure.
Visa Fall 2026 Biannual Threats Report. Incidents involving third-party agents rose 62% on the previous six months.
These figures come from different datasets and should not be combined into one growth rate.
The figures do not establish that attacks on core financial systems have stopped. Verizon still lists system intrusion, ransomware, vulnerability exploitation and stolen credentials among the major threats to the sector. What the research does show is that the attack surface extends well beyond the systems financial institutions own and administer.
I am also wary of a conclusion that appears whenever cyberattack numbers rise. More attacks do not automatically mean security is getting worse. Visa reported that payment account enumeration attempts increased by 35% year over year, while fraud losses associated with enumerated credentials fell by 21.4%. Better detection and intervention made a difference, and that is evidence defensive investment works.
What I want to know is whether those investments are being assessed against the right outcomes. If a security programme prevents an attack on a database but cannot detect the misuse of an authorised payment integration, what does its success tell us about the institution’s financial exposure? We may be drawing conclusions from measurements that cover only part of the business.
The Comfort of a Green Dashboard
Imagine a board meeting at a bank. The CISO presents the quarterly cybersecurity report. Vulnerability remediation is on target, patch compliance is improving, endpoint protection coverage is satisfactory and the latest audit has produced no critical findings. Management has plans for the remaining risks.
There is nothing wrong with that report, and I would expect a competent security programme to provide it. I would also want to know whether the institution understands the relationships through which an attacker could reach its most important business functions.
| What the report shows | What I would also want to know |
|---|---|
| Vulnerability remediation is on target | Is there an attack path through legitimate credentials, a trusted integration or an externally managed application? |
| The penetration test produced no critical findings | Which important dependencies were outside its scope? |
| Patch compliance is improving | If a security gateway is compromised, what stops the attacker reaching systems beyond it? |
| The latest audit is clean | What happens when a trusted third-party JavaScript file starts serving malicious code, and how would we detect a compromise in our software build pipeline? |
| Vendors have passed their assessments | Which external identities and service accounts can influence financial transactions? |
I am not suggesting that vulnerability management, penetration testing or compliance assessments are ineffective. I have spent too many years implementing and reviewing these programmes to dismiss their value.
I am concerned about treating the results as proof that the institution is secure without examining what was covered.

A meme making the rounds on Twitter.
There is a regulatory dimension too. The Reserve Bank of India’s directions on outsourcing of information technology services recognise that regulated entities remain responsible for oversight and risk management even when external providers perform the activities. Europe’s Digital Operational Resilience Act addresses third-party ICT risk, testing and concentration risk across the financial sector. These are necessary developments, but an institution can satisfy vendor assessment requirements and still not understand how a compromised provider could affect its operations. A questionnaire can tell us that a provider has access controls and performs security testing. It may not tell us whether an attacker using that provider’s privileges could initiate a transaction, extract customer information or alter a production application. I would rather know the answer to those questions than receive another assurance that the provider has passed its annual audit.
AI Will Make These Problems Harder
Artificial intelligence is part of this discussion now, although I think we need to be careful about what the evidence establishes. The Binance investigation described a malicious installer linked to an AI coding tool and the compromise of an internal AI tool during the attack. That does not establish that an autonomous AI system planned or carried out the theft. It does show that AI tools, like other development and operational tools, can become part of an attack path, so their identities, credentials, execution privileges and access to other systems need to be understood.
There is also evidence of attackers using AI to improve phishing, impersonation, reconnaissance and malicious software development. Visa’s November report describes criminals applying AI to existing fraud techniques to increase speed and scale.
I want to make one observation here. AI is being used effectively to identify vulnerabilities and, in some cases, to chain attacks and even reason out the full attack cycle. It is still not the predominant attack vector threat actors use. Attacks are largely driven by human social engineering, automation and a lot of preparation, and the Binance campaign fits that pattern. The overwhelming sensational news that AI will be used for attacks at machine speed is unproven so far. On Venus, our deception platform for observing real-world threats, we see less than 2% of suspected agentic AI threats.
AI is not the predominant attack vector. Attacks are still largely driven by human social engineering, automation and a lot of preparation.
I have been writing about agentic reconnaissance and its implications for cybersecurity for more than a year. My concern is that attackers could use systems capable of examining an environment, investigating relationships, testing access paths and adapting with limited human intervention. Consider an institution with hundreds of applications, thousands of identities, multiple service providers and software dependencies that change every week. An attacker may be able to investigate those relationships faster than the institution can review them through its usual assessment cycles. We already struggle to keep accurate inventories of the systems we own, and understanding the relationships between systems, service providers and business processes is harder.
AI did not create this problem. It can make the consequences worse.
What I Would Ask in the Boardroom
If I were reviewing the cybersecurity posture of a financial institution, I would want to understand its critical business processes before studying the security dashboard.
Three places I would start
The movement of money
Which applications, identities, service providers and integrations can influence a transaction? Which can initiate payments, change details or authorisations, or deploy software that affects financial processing? How do we know those permissions cannot be misused?
The technology dependencies
What third-party code runs in customer-facing applications? Which external services can introduce changes without a deployment by our own team? Do we monitor those changes and restrict what the code can execute or transmit?
Containment
If a service provider is compromised, can we revoke its access without disrupting unrelated operations? If a developer’s credentials are stolen, can the attacker deploy to production? If an authorisation has been set up outside the normal transaction channel, does suspending that channel stop the activity?
The Optimizely incident gives us a reason to ask the second set of questions, and so do the npm compromises. The relevant attack path might sit in a browser, on a developer’s computer or inside a software build pipeline, long before anything unusual appears in a core banking system. The Binance incident shows why the last question matters. Its normal withdrawal infrastructure could be suspended while malicious blockchain authorisations stayed effective. These questions need to be tested. A risk register or a statement that controls have been implemented cannot answer them.
There is also the question of responsibility. Cybersecurity teams can explain the controls protecting an application, but understanding what the application means to the business takes operations, finance, technology and risk management. The CISO should not be expected to understand every business dependency without the people who run those processes. Equally, business leadership should not assume the security team has assessed every relationship because an application appears on an asset inventory.
We have spent years encouraging organisations to move cybersecurity into the boardroom, and that was necessary. Bringing security reports into the boardroom is not enough, though. The discussion has to reflect how the business operates and where its dependencies lie.
The Financial System We Have Built
I do not believe financial institutions are necessarily less secure than they were ten years ago. Many have invested in security capabilities, improved detection, strengthened authentication and developed incident response processes that can contain attacks before they cause greater damage.
The financial system has changed during the same period. Money moves faster. Transactions depend on more organisations. Applications contain software written and maintained by people the institution has never met. Customer-facing services rely on scripts delivered from external infrastructure. Privileges extend across cloud services, development platforms, identity providers and payment-processing systems. The boundaries of the business have expanded, and I am not convinced our approach to measuring security has kept pace.
| Incident | Where it began | What it reached | Status |
|---|---|---|---|
| Central Bank of India | Outward RTGS system, alleged SFMS access | 187 transfers from the RBI settlement account | Cause not established publicly |
| State Bank of India | Not yet established | Customer information on the dark web | Independent investigation under way |
| PHP2Shell | Internet-facing security gateway portals | A VMware ESXi host and destructive malware, in one intrusion | Exploitation confirmed |
| npm packages | Malicious package versions installed by developers and build systems | Developer and cloud credentials | Reported by Microsoft |
| Optimizely | A flaw in account signup | Visitors of 178 organisations | Closed by Optimizely. Full extent unknown |
| Binance | Engineers’ devices and tools | Wallet-signing operations | Preliminary investigation |
| Crédit Mutuel Arkéa | An external service provider’s platform | Fraudulent transfers from the bank’s own account | Contained, per the bank |
There is still a common lesson. The systems we regard as central to a business are not necessarily where an attacker needs to begin. That is why I do not think securing core systems alone is an adequate measure of financial security. It remains essential, but we also need to understand the infrastructure, code, identities and relationships through which those systems and their customers can be reached.
We cannot remove these dependencies without dismantling much of what makes modern finance possible, and we should not try. Outsourcing, digital platforms and interconnected payment systems have made financial services faster and open to more people. What we can do is understand the dependencies, limit the authority they carry, observe what they do and test what happens when one of them is compromised.
A bank can be compliant, run a mature cybersecurity programme and report strong security metrics. Those achievements matter. They do not establish that every path to its money, information and business processes has been understood or controlled. We have built a financial system around interdependence. Attackers have learned to exploit the connections between its components, and those connections are not always visible in the reports that reach management.
The closing signal
If we continue to measure security within the boundaries of individual systems while financial risk moves across them, we will keep missing part of the problem.
And that, to me, is the system of financial insecurity.
Sources and Further Reading
- Central Bank of India, Stock exchange disclosure on the RTGS issue, 19 October 2026; MediaNama, Central Bank of India alleges hackers siphoned off Rs 13.97 crore through financial messaging system. The ₹13.97 crore figure comes from the police complaint as reported, not from the bank’s disclosure.
- Hindustan Times, State Bank of India data breach; BazaarWatch, reproduction of the 21 October stock exchange disclosure. SBI’s figures are preliminary.
- Unit 42, Threat Brief: Critical Vulnerabilities in PHP; CISA, Critical Vulnerabilities in PHP Require Immediate Patching.
- Check Point Research, PHP2Shell exploitation on Check Point gateways, 12 November 2026.
- Cyber Security Agency of Singapore, Advisory AL-2026-157, 3 November 2026.
- Microsoft Security, ChainDrop supply-chain compromise, 4 August 2026.
- Report URI, World’s largest brands infected with malware via Optimizely; Optimizely, Security Incident Report, 17 November 2026; Optimizely, incident status history.
- Binance, Investigation report, 3 November 2026; Binance, Security incident announcement; The Hacker News, Binance suffers security breach, 30 October 2026.
- FBI, North Korea responsible for $6 billion Binance attack, 31 October 2026.
- Crédit Mutuel Arkéa, Press release, 4 November 2026, and follow-up questions and answers.
- Verizon, 2026 Data Breach Investigations Report; Visa, AI-enabled fraud and third-party attacks are scaling worldwide, 17 November 2026.
- Reserve Bank of India, Master Direction on Outsourcing of Information Technology Services, 10 April 2023; European Banking Authority, Digital Operational Resilience Act.
