I. Introduction

Thank you for that kind introduction, Neil [Strauss, VP and Senior Credit Officer, Moody’s Investors Service], and thank you for inviting me to speak with you and your colleagues at Moody’s here today.

We are living through a transformation, as the world shifts from analog to digital. Economies, markets, commercial enterprises, and governments are becoming more data-driven and technology-dependent. Companies increasingly are becoming more intertwined with technology, restructuring their operations and businesses accordingly. What is clear is that technology continues to offer the promise of a more modern world with interconnected economies and communication and financial systems.

Now the technologies of artificial intelligence are being added into this transformation. AI has the potential to accelerate the diagnosis of disease, help combat climate change and streamline manufacturing processes. Most large companies already use AI in some form in their operations and financial reporting. It helps enhance efficiency and accuracy, and boosts productivity.

But real perils exist, too. Generative AI models, which create new data in response to queries, can hallucinate or make up content. AI can accelerate the spread of disinformation and the erosion of privacy. It presents risks to cybersecurity, as well.

To many of us in this room today, it may seem like “déjà vu all over again” as Yogi Berra famously quipped … because these are some of the same risk-reward tradeoffs we faced through earlier stages of the move from analog to digital. And, at the same time, it’s exponentially different in its breadth and speed of change.

Over the next thirty minutes or so, I’d like to share my thoughts on the current state of cybersecurity and how AI is affecting the threat and response environment, in particular. I’ll also touch on public policy challenges AI presents to cybersecurity and how governments and regulators are responding. Global cooperation remains essential.

II. Current Environment

A. Nature of Cyber Threats

Let’s consider the cyber threat environment in which we currently live. Threat actors around the world continue to exploit the Internet to engage in damaging mischief, criminal activity, and geopolitical disruption. Headlines report the infiltration of computer systems at hospitals and healthcare providers, financial institutions, technology companies, and government agencies. Operations disrupted and patients put at risk. Personal and other sensitive information exposed. Business activities disrupted. Money stolen. Trust in institutions undermined.

Now some attackers are adding machine learning and AI to their bags of tricks. Machine learning teaches computers how to automatically perform specified tasks and quickly discern patterns. Artificial intelligence goes a step further. Computers are used to simulate the processes around human intelligence, as if they can think, make sense of huge amounts of data, and learn from experience.

The term “AI” covers a broad swath of applications that mimic human neural networks. It includes chatbots – or computer programs that simulate written or oral human conversations – as well as deep-learning and large-language models.

Generative AI models – such as Microsoft’s Copilot and OpenAI’s ChatGPT – have been trained on vast amounts of data accessed through the Internet. These models can be used to compose emails, write computer code, and translate languages. Beyond prose, generative AI models can create digital images and videos and audio recordings.

To defend against cybersecurity threats, the largest companies are also deploying machine learning and AI. Using sophisticated algorithms, AI-based systems can analyze millions of pieces of data, detect malicious software (or malware), run pattern recognition scenarios, and disarm identified malware, all before it ever enters a company’s networks.

Given the ready availability of many AI tools, however, the barriers to entry have been lowered for novice cyber criminals, hackers-for-hire, and hacktivists to engage in ever more sophisticated attacks.

For some insights on how AI is starting to change the landscape, let’s dive deeper into three of the most common and costliest types of cyberattacks – ransomware, business email compromises, and stolen credentials.

1. Ransomware

Last November, the US arm of the world’s largest lender by assets, Industrial and Commercial Bank of China (ICBC), experienced a large-scale ransomware attack. This type of attack involves criminals accessing computer systems and deploying malware to paralyze those systems, then demanding ransom, typically in the form of virtual currency, for the electronic keys to release the locked items.

In this case, the bank’s US broker-dealer was the target. That entity operates from New York and its business is primarily the clearing and settling of US Treasury transactions; basically, ensuring the timely exchange of money and securities for previously agreed-to trades.

With corporate emails and other computer systems locked, bank personnel resorted to personal Gmail accounts and flash drives to process trades. Bank clients and partners also had to reroute transactions to other institutions. The bank was forced to inject capital into its US broker-dealer to settle trades and repay debts. Before it was all over, ICBC owed BNY Mellon a balance of $9 billion – an amount that was greater than its broker-dealer’s net capital.

Ransomware (or similar extortion-type attacks) are a top cyber threat across almost all industries.[1] The average cost of a ransomware breach – not including any ransom paid – has reached $4.91 million, according to a recent global study by IBM.[2] But total costs are likely to be exponentially higher when lost revenue is considered.

Last year’s cyber-incident at consumer-product giant The Clorox Company provides some useful insights. While the company has never publicly disclosed the exact nature of the attack, experts have opined that it had all the hallmarks of ransomware.

What do we know? Quite a bit because of rules adopted last year by the Securities and Exchange Commission (SEC).[3] Those rules require US public companies to disclose material cybersecurity incidents within four business days of discovery.[4] They also mandate that public companies estimate and disclose the costs associated with material cybersecurity incidents.[5]

Clorox initially disclosed that it had identified “unauthorized activity” on some of its information technology systems.[6] The company said the intrusion caused large-scale disruption of its operations, triggering significant product shortages and order processing delays. The company attributed a quarterly sales decline of 20 percent, or over $350 million, to the event. Beyond lost revenue, the company pinned the cost of its response and recovery efforts at $59 million, of which only $30 million was covered by insurance.  

As disruptive as ransomware can be, business email compromises and stolen credentials can be even more dangerous. This is because intruders typically remain undetected longer in computer systems and networks, giving them time to find and exploit victims’ vulnerabilities then cover their tracks.

2. Business Email Compromises

Business email compromises are scams where criminals use manipulated – or spoofed – email addresses and domains to impersonate company executives to dupe employees into making unauthorized payments. Unlike ransomware, victims typically do not initially know that they have been tricked. By the time the deception is discovered, the money is usually long gone.

Attackers can now use generative AI to better impersonate company executives. With these tools, emails can be drafted not only to mimic the style and tone of identified executives, but the spelling and grammar will likely be impeccable, too. So, detection becomes ever more difficult.

While only accounting for 10 percent of data breaches, business email compromises are the second most expensive, costing on average almost $4.9 million to remediate, according to IBM.[7] The annual aggregate costs to businesses, however, likely runs in the tens of billions of dollars or more.[8]

A critical infrastructure company operating out of New York had a recent harrowing experience.[9] A compromised email duped the company into transferring $50 million into bank accounts controlled by criminals. The company discovered the fraud while most of the payments were still being processed and contacted the FBI. The FBI sent a “Financial Fraud Kill Chain” request to the financial institutions that had initially received the funds. Given that quick action, $46 million of the funds were frozen in place.

As to the probability of recovering the remaining funds, we know that the likelihood declines precipitously once the money has left the control of the receiving financial institutions. We also know that recovery efforts are time consuming and costly, and historically have had limited success.

3. Stolen Credentials

Lastly, stolen credentials continue to be a common and costly attack vector. According to IBM, compromised credentials were responsible for 16 percent of the data breaches analyzed in their latest study, costing on average just over $4.8 million.[10]

Recall a decade ago when compromised credentials were believed to have been used in a series of cyber-enabled thefts using the international payment messaging system operated by SWIFT [Society for Worldwide Interbank Financial Telecommunication]. The most well-known of these thefts involved the central bank of Bangladesh. Criminals masquerading as authorized users attempted to steal $1 billion from the bank. The crooks ended up walking away with $81 million, most of which was never recovered.

Today there are new ways to steal credentials. One way is through a voice phishing attack, known as “vishing.” Instead of emails or texts, perpetrators combine phone calls and voice messages with social engineering to convince victims to reveal sensitive information, such as usernames and passwords. Media reports have pointed to vishing of an IT help desk as the original source of last year’s ransomware attack against the gaming and hospitality company, MGM Resorts International. MGM has publicly pegged the cost of that attack at $110 million.[11]

Now think about the effects of combining vishing with recently available voice-cloning technology. These online AI-powered tools allow almost anyone to upload a few seconds of a person’s voice, type in what they want that voice to say, and quickly create a hyper-realistic digital recording or deepfake.

Audio deepfakes match so closely the tone, pitch, and mannerisms of the input voice that unsuspecting listeners may not detect the deception. All for a monthly subscription fee of $5 to the AI software provider. Think about the potential havoc caused by deepfake voices of company CEOs or CFOs.

B. Effect of Artificial Intelligence

Since the public introduction of generative AI almost two years ago, cybersecurity experts have been watching for a potential shift in the threat landscape. Deepfake technology has advanced and been tied to reported fraud and misinformation campaigns, at least anecdotally.[12] Evidence also exists that known state-sponsored threat actors are experimenting with generative AI, using it as a job aid to, for example, help write and debug computer code for more effective attacks.[13]

On the defender side of the equation, the use of security AI, machine learning, and other automation techniques appear to be making a difference.[14] These technologies have helped organizations identify and contain breaches and reduce costs. When AI is deployed extensively across operational workflows focused on preventing attacks, organizations averaged $2.2 million less in breach costs compared with entities not similarly using AI, according to IBM; this amounted to an almost 46 percent cost savings differential.

III. How to Respond?

Given the current state of cybersecurity and AI, how do organizations, governments, and regulators respond?

A. Embracing Effective Corporate Governance

We know that implementing cybersecurity frameworks that explicitly incorporate governance and oversight make a difference. The National Institute of Standards and Technology (NIST) has a cybersecurity framework that entities around the globe have been using since its creation in 2014. One of its original shortcomings, though, was that it did not explicitly recognize governance as part of its framework.

That flaw was rectified earlier this year when NIST published the 2.0 version of its framework.[15] The updated framework has added governance to five other key functions: identify, protect, detect, respond, and recover. With that addition, the framework provides a comprehensive approach for managing cybersecurity risk, one that makes holistic oversight paramount. The updated framework also makes clear that it applies to all types of technology environments and explicitly references AI systems.

B. Identifying and Implementing Baseline Protection

We also know that the most effective cybersecurity frameworks incorporate basic cyber hygiene as essential components. According to an analysis by IBM, many of the factors that reduce the average cost of a data breach are basic cyber hygiene, including employee training, identity and access management controls, and incident response planning.[16] We also know that each year entities, large and small, have significant cyber incidents because they fail to get the basics right.

But having baseline practices in place may no longer suffice to keep regulators at bay. The SEC recently brought enforcement actions alleging, among other charges, that two public companies that were victims of ransomware attacks failed to have effective cybersecurity controls in place. For one company – R.R. Donnelley & Sons Co. – the alleged deficiencies related to its incident response plan. For the other company – SolarWinds Corp. – the alleged deficiencies related to poor system identity and access controls.

The SEC alleged that the deficiencies at the two companies amounted to internal accounting control failures that were also statutory violations. In both actions, the SEC applied in a novel way the securities laws that require public companies to have appropriate internal accounting controls.[17]

R.R. Donnelley settled with the SEC, paying a civil money penalty of just over $2 million.[18] SolarWinds Corp. is litigating with the SEC. This past summer, the US district court dismissed much of the SEC’s complaint, rejecting the agency’s attempt to leverage internal accounting control provisions to regulate general cybersecurity practices at public companies.[19]

The court ruled that the SEC’s authority to regulate a public company’s systems of internal accounting controls does not extend to overall corporate cybersecurity practices but instead is limited to the company’s financial accounting controls. Nonetheless the SEC’s increasingly aggressive stance has companies evaluating whether their cybersecurity practices are comprehensive and effective, and whether those practices follow other SEC cybersecurity rules, including those requiring prompt disclosure of material cyber incidents.

All of this suggests that organizations may need even more robust cybersecurity protections. Advanced organizations are moving toward zero-trust models that assume user identities or networks themselves are already compromised.

Here is where AI can be transformational. Continuously scanning systems to validate users’ connections and activities, looking for anomalies. Other AI-powered tools can help by automatically correlating data from different sources and instantaneously interacting with a company’s security software to remediate incidents in real time. All with little or no human intervention.

But what are smaller or less advanced organizations and countries to do?

C. Cooperating Internationally

To varying degrees, individual governments, regulators, and companies have learned that they cannot go it alone when it comes to cyber. But given the scope and scale of the threat and our global interdependencies, more needs to be done to work together.

For example, countries have been more successful at protecting their financial systems and sectors when they coordinate with like-minded countries. Why? Malicious actors often use the same or similar methods to target multiple institutions across jurisdictions. That is why sharing information about cyber vulnerabilities, threats, and incidents is key.

We see some of this international cooperation starting to happen around AI cybersecurity. For example, late last year, Britain convened an international summit to develop global safety standards for AI. Representatives from 28 countries attended and signed a communique committing to future cooperation.[20] Also, late last year, President Biden issued an executive order directing federal government agencies to take vital steps to better ensure safe, secure, and trustworthy AI.[21] The executive order identified 21 countries with which the US government had consulted. Both the communique and the executive order noted the significant perils that AI presents to cybersecurity.

Beyond that, the United Nations released a report this month on international governance of AI.[22] The report provided a blueprint for addressing AI-related risks while better ensuring its transformational potential is shared globally. To date only a handful of countries have laws governing the use of AI. The US, characteristically, has taken a wait-and-see approach, relying instead on the bully pulpit and voluntary compliance.

IV. Conclusion

Governments and public and private organizations across the globe are fighting on all fronts to combat rising cyber dangers as the world moves towards a more interconnected digital landscape – a landscape that is just starting to recognize the promise and threat that AI presents to cybersecurity. What is clear, though, is that increasingly cybersecurity and AI are both critical and interdependent.

Thank you for your attention.

I’m happy to take your questions.