AI-Enabled Cyberattacks: Emerging Legal Standards for Corporate Cybersecurity Programs

Cyberattacks were already expensive and frequent before AI. They have become cheaper, faster, and harder to detect since. The same AI tools that help businesses automate customer service, generate marketing copy, and analyze financial data are being used by threat actors to write more convincing phishing emails, generate deepfake audio of your CEO authorizing a wire transfer, identify exploitable vulnerabilities in your infrastructure at scale, and conduct attacks that previously required a team of sophisticated adversaries working for weeks.

For corporate legal and compliance teams, the question is not whether AI is changing the threat landscape — it is. The question is how regulators and courts are responding, and what the evolving legal standard for “reasonable” cybersecurity looks like in an environment where AI is amplifying every attack vector.

The answer, emerging clearly in 2025 and 2026 guidance and enforcement, is that the bar for adequate corporate cybersecurity is moving upward. The New York Department of Financial Services issued its most specific AI-focused cybersecurity guidance to date in May 2026. The FTC and SEC have both intensified scrutiny of how companies document and manage AI-related cybersecurity risks. California’s audit and risk assessment mandates apply to a broad range of businesses. And what counts as “reasonable” security under FTC enforcement is increasingly being defined by reference to whether companies have specifically addressed AI-enabled attack methods.

This post covers where those legal standards stand today and what an AI-resilient cybersecurity program looks like in practice.

The Threat Landscape: Why AI Changes the Calculus

Understanding the regulatory response requires understanding what AI has actually changed about cyberattacks. The threats are not new in kind — phishing, social engineering, vulnerability exploitation, ransomware, and business email compromise have existed for years. What AI has changed is the economics and scale.

AI-powered phishing and social engineering. Traditional phishing was constrained by the cost of personalization. Mass phishing campaigns were easy to spot because they were generic. Targeted “spear phishing” required human research and was expensive to conduct at scale. AI changes this: large language models can generate highly personalized, grammatically perfect phishing emails in volume, tailored to the specific target using information scraped from LinkedIn, corporate websites, and social media. The unit cost of a convincing personalized attack has collapsed.

Deepfake audio and video for business email compromise. The FBI has documented a surge in what it calls “voice cloning” attacks, where threat actors generate realistic audio of a company executive using AI voice synthesis, then call employees and instruct them to authorize wire transfers or provide credentials. NYDFS’s May 2026 guidance specifically called out deepfake audio and video as threats that undermine traditional authentication methods. Attacks that previously required insider access or elaborate impersonation schemes can now be executed with publicly available audio samples and AI tools.

AI-accelerated vulnerability discovery. Security researchers and threat actors alike are using AI to analyze code and infrastructure for exploitable weaknesses at a speed that was previously impossible. A threat actor using AI-assisted vulnerability scanning can identify and begin exploiting a newly disclosed software vulnerability before most companies have even finished reading the disclosure notice, let alone patched the affected system.

AI-generated malware. Security researchers have demonstrated that AI systems can generate functional malware variants that evade signature-based detection. Each variant is slightly different from known malware, allowing it to bypass traditional endpoint detection tools that rely on matching known patterns.

The cumulative effect is that the window between when a threat actor identifies a target and when they execute a successful attack has shortened, and the sophistication required to execute a credible attack has dropped. Both of these shifts have implications for what a “reasonable” cybersecurity program must include.

The NYDFS Framework: The Most Specific AI-Cybersecurity Guidance in the US

The New York Department of Financial Services regulates banks, insurers, money transmitters, mortgage companies, and other financial services businesses doing business in New York. Its cybersecurity regulation, 23 NYCRR Part 500, is widely regarded as the most comprehensive state-level cybersecurity mandate in the United States, and NYDFS enforcement actions have produced the largest state-level cybersecurity penalties imposed anywhere in the country.

On May 21, 2026, NYDFS issued two Industry Letters that represent the clearest regulatory statement in the US to date on what AI-resilient cybersecurity requires:

Industry Letter 1: “Heightened Cybersecurity Risks Associated with Frontier AI Models.” This advisory identifies frontier AI models — defined as AI models that amplify the potency, scale, and speed of identifying vulnerabilities and exploits in information systems — as a distinct category of threat requiring specific mitigation. The Advisory directs CISOs at covered entities to assess how AI-enabled attack capabilities affect their specific risk profile and to implement countermeasures accordingly.

Industry Letter 2: “Guidance on Measures Regulated Entities Should Consider in a Heightened Cybersecurity Threat Environment.” This guidance catalogs specific security measures across three categories: (1) reducing the attack surface, (2) improving threat detection and readiness, and (3) strengthening resilience and response.

While both documents state that they do not alter the formal requirements of Part 500, NYDFS has made clear through its examination process that these publications serve as reference points. A covered entity that has not considered these measures — and that cannot document why it adopted or declined them — faces examination risk even if every formal Part 500 requirement is technically satisfied.

AI-Specific Recommendations from NYDFS

The NYDFS guidance makes specific recommendations about authentication and access controls that respond directly to AI-enabled attack techniques:

Phishing-resistant multi-factor authentication. NYDFS recommends that covered entities implement authentication methods that AI cannot defeat. Specifically, the guidance recommends digital-based certificates and physical security keys as authentication methods for access to systems holding sensitive data. The reason: AI-generated deepfakes can defeat knowledge-based authentication (voice recognition, security questions) and can generate convincing simulations that fool employees into providing SMS codes. Hardware security keys and certificate-based authentication are resistant to these attacks because they require physical possession of a device, not just knowledge of a credential.

Third-party and vendor AI risk. NYDFS explicitly requires covered entities to assess AI-related risks from third-party vendors. If a vendor is using AI to manage or process data that the covered entity is responsible for protecting, the covered entity must evaluate whether the vendor’s AI systems create additional attack surfaces — and whether AI attacks against the vendor’s systems could compromise the covered entity’s data. This extends the supply chain security obligation into the AI context.

Incident response planning for AI-enabled attacks. NYDFS recommends that incident response plans specifically address scenarios involving AI-enabled phishing, deepfake social engineering, and AI-accelerated exploitation of known vulnerabilities. Generic incident response plans that were written before these attack types existed may not adequately address the response requirements for AI-enabled incidents.

The FTC’s “Reasonable Security” Standard and AI

The FTC enforces data security requirements for a broad range of businesses under Section 5 of the FTC Act, which prohibits unfair or deceptive practices. The FTC has pursued dozens of enforcement actions against companies for inadequate data security, and its enforcement record has progressively defined what “reasonable” security requires.

The FTC’s Office of Technology and the Division of Privacy and Identity Protection published recommendations in December 2024 specifically addressing security risks in AI development. The FTC’s position: companies developing AI systems must address cybersecurity risks in the development and deployment of those systems, not just in their general IT infrastructure.

What the FTC’s enforcement record tells us about the evolving standard:

• Failure to implement multi-factor authentication has been a recurring finding in FTC data security enforcement actions. Post-2025, the question is not whether MFA is implemented but whether the MFA is AI-resistant.
• Failure to patch known vulnerabilities within a reasonable timeframe is a consistent FTC finding. The accelerated exploitation timeline created by AI-assisted vulnerability scanning compresses what constitutes a “reasonable” patching window.
• Failure to adequately manage third-party vendor access is a consistent finding. AI-powered vendors create additional surface area that must be addressed in vendor risk management programs.
• Failure to train employees on current social engineering techniques is a consistent finding. AI-generated phishing and deepfake business email compromise require employee training that addresses these specific techniques.

The FTC has not issued specific AI cybersecurity rules for general commercial businesses outside of sector-specific contexts. But its enforcement trajectory — combined with the NYDFS guidance, which tends to influence other regulators — points clearly toward a standard that accounts for AI-enabled attack methods.

California’s Cybersecurity Audit and Risk Assessment Mandates

California implemented cybersecurity audit and risk assessment requirements for covered businesses under its California Consumer Privacy Act / California Privacy Rights Act framework, with requirements phasing in beginning January 1, 2026.

Cybersecurity audits. Businesses that process personal information of California consumers and whose processing activities present significant risks to consumers’ security must conduct annual cybersecurity audits by qualified auditors. The audit must assess whether the business’s security practices meet a “reasonable security” standard.

Risk assessments. Covered businesses must conduct and document risk assessments before initiating processing activities that present heightened risk to consumer security. These assessments must identify and evaluate security risks, including risks from third-party systems.

What AI means for California risk assessments. If your business uses AI systems that process California consumer data — an AI-powered customer service tool, an AI recommendation engine, an AI fraud detection system — those systems need to be assessed as part of your risk assessment obligations. The AI vendor’s security practices are part of the assessment. AI-enabled attack methods (phishing, deepfake social engineering, AI-accelerated vulnerability exploitation) are part of the threat landscape that the risk assessment must address.

California’s audit and risk assessment requirements are enforced by the California Privacy Protection Agency, which began exercising its enforcement authority in 2024. The penalties for violations can reach $7,500 per intentional violation, with class action exposure under separate CCPA provisions for security breaches resulting from failure to maintain reasonable security.

What “Reasonable Security” Means Post-2026: A Practical Standard

Synthesizing the NYDFS guidance, FTC enforcement record, California requirements, and NIST Cybersecurity Framework, a post-2026 “reasonable security” standard for businesses that process significant volumes of personal or financial data includes the following elements:

1. AI-resistant authentication. Passwords alone are no longer reasonable security for access to sensitive systems. MFA is the minimum. Post-2025, MFA that can be defeated by AI (SMS codes that can be phished by AI-generated deepfake calls, knowledge-based authentication that can be defeated by AI-researched answers) is increasingly inadequate for high-risk systems. Hardware security keys or certificate-based authentication is the direction of travel for access to the most sensitive systems.

2. AI-enabled threat detection. Using AI to defend against AI-enabled attacks is increasingly the baseline. Behavioral analytics, anomaly detection, and AI-assisted log analysis are needed to detect AI-driven attacks that bypass traditional signature-based tools. The NYDFS guidance specifically recommends enhanced threat detection capabilities for frontier AI attack scenarios.

3. Reduced and documented attack surface. The NYDFS guidance emphasizes “reducing the attack surface” as a primary security objective. This means minimizing external-facing systems, disabling unused services and ports, segmenting networks to limit lateral movement after a breach, and maintaining an accurate and current inventory of all systems, applications, and vendors. For AI specifically: every AI vendor integration is an additional attack surface point that must be inventoried and managed.

4. Written incident response plan that addresses AI attack scenarios. A generic incident response plan that was written without reference to AI-enabled attacks is likely inadequate in a post-2025 regulatory environment. The plan should address: what to do when an executive’s voice is deepfaked to authorize a fraudulent transaction; how to respond to an AI-accelerated compromise of a known vulnerability; how to detect and respond to phishing campaigns that defeat conventional spam filters.

5. Regular employee training on AI-enabled social engineering. Employees are the first line of defense against phishing and business email compromise, and training must keep pace with the current threat. Training that addresses only legacy phishing techniques (generic emails from Nigerian princes) is inadequate. Employees who handle financial transactions or have access to sensitive systems need specific training on deepfake audio and video attacks, AI-generated spear phishing, and verification protocols for high-risk requests.

6. Comprehensive vendor risk management that includes AI vendors. If you use AI vendors — and almost every technology-dependent business now does — you need to assess those vendors’ security posture with the same rigor you apply to other critical vendors. Your vendor due diligence questionnaire needs to address AI-specific risks: Does the vendor use AI to manage or process your data? What is the vendor’s posture on prompt injection and adversarial attacks against their AI systems? How does the vendor detect and respond to AI-enabled attacks against its own infrastructure?

7. Documentation of all of the above. Regulators investigating a breach want to see evidence that you had a reasonable security program, not just a claim that you did. Written policies, risk assessments, audit results, training completion records, and vendor due diligence documentation are the evidence that demonstrates reasonable security. Companies without this documentation have no defense when a breach occurs.

SEC Cybersecurity Disclosure Requirements

Public companies face an additional layer of AI-related cybersecurity obligation under the SEC’s cybersecurity disclosure rules, which took effect in 2023. Public companies must:

• Disclose material cybersecurity incidents within four business days of determining that the incident is material;
• Disclose annually: the board’s oversight of cybersecurity risk, management’s role in assessing and managing cybersecurity risk, and the company’s cybersecurity risk management processes.

For public technology companies, the AI-cybersecurity overlap is a disclosure risk as well as an operational risk. If your company’s AI systems are a significant part of your business and those systems face material cybersecurity threats — including the AI-specific threats described in this post — that risk may be material and subject to disclosure. The SEC has brought enforcement actions against public companies for inadequate cybersecurity risk disclosures, with penalties reaching into eight figures.

Practical Takeaways for Businesses

Assess your AI vendor exposure now. List every AI tool and vendor your company uses. For each, assess: What data does this system access? How is that data secured? What happens if the AI vendor’s system is compromised? What is the vendor’s security track record?

Update your threat model for AI-enabled attacks. Your risk assessment should specifically identify AI-enabled phishing, deepfake social engineering, and AI-accelerated vulnerability exploitation as current threats. These are not hypothetical future risks — they are documented current attack methods.

Upgrade authentication for high-risk systems. At minimum, implement phishing-resistant MFA for all systems that hold sensitive financial or personal data. For the most sensitive systems — financial transfer authorization, access to systems holding large quantities of personal information — evaluate hardware security key deployment.

Train your employees for the current threat environment. AI-generated phishing and deepfake business email compromise require specific, current training. Include a call-back verification protocol for high-risk financial requests: any request to transfer funds, change payee information, or provide credentials received via email or phone call should be verified through an out-of-band channel before action is taken.

Review your incident response plan. Confirm that your plan addresses AI-enabled attack scenarios and that the plan has been tested in the last 12 months through a tabletop exercise or simulation.

Document everything. Write down your risk assessments, your vendor evaluations, your training program, and your security decisions. The documentation is evidence of reasonable security — and a good-faith effort to address AI-enabled threats is, right now, the leading edge of what regulators are looking for.

This post is for general informational purposes only and does not constitute legal advice. Reading this post does not create an attorney-client relationship. If you have questions about your specific situation, consult a qualified attorney.