The advent of frontier AI models that are both increasingly adept at identifying software vulnerabilities and developing exploits, and at defensive activities such as patching, has led to significant discussion about the potential implications for the cybersecurity risk landscape. The incorporation of AI-enabled capabilities as part of a company’s defensive posture – and as part of the threat actor toolkit – is not new. Still, the recent expansion in cyber-related capabilities of the latest generation of AI models, including Anthropic’s Mythos model and OpenAI’s GPT 5.4-Cyber, has the potential to significantly reduce the expertise and time required to execute intrusions, including advanced intrusions that were previously only within the ability of the most sophisticated threat actors.
Efforts to address these emerging risks are gaining significant momentum. Regulators and other government agencies – already focused on managing AI-enabled cyber risks – are now emphasizing the heightened risks posed by frontier AI models, such as the May 21, 2026 Industry Letter to regulated entities published by the New York State Department of Financial Services (“NYDFS”). More recently, on June 2, 2026, President Trump signed an Executive Order on Promoting Advanced Artificial Intelligence Innovation and Security (“AI EO”), which will establish a framework for secure development of frontier AI models and an “AI cybersecurity clearinghouse” to facilitate vulnerability coordination and remediation, among other initiatives. The AI EO was quickly followed by National Security Presidential Memorandum 11, published on June 5, 2026, which directs the military, intelligence agencies, and related federal departments to accelerate the adoption of AI for national security applications.
These policy developments, along with the recent announcement by Anthropic that it would expand “Project Glasswing” to provide early access to its Mythos model to approximately 150 additional organizations, underscore the importance for lawyers and their clients to evaluate the implications of this changing landscape of cybersecurity risks. Against this fast-moving and evolving backdrop, organizations will likely face new risks and challenges as they consider how to respond to and protect against threats operating at the speed and scale of advanced AI. However, foundational governance principles and underlying controls continue to be important risk mitigators for organizations as they evolve with the new cybersecurity landscape – whether by hardening their systems or planning to incorporate AI capabilities (including the use of agentic AI) into their cyber defenses. This alert identifies practical actions legal teams can take to help their organizations address the risks and embrace the opportunities presented by these models.
Assess whether existing risk management approaches should adjust in response to AI-related cyber risks: Frontier AI models can accelerate and scale vulnerability discovery and security analysis, which can compress decision timeframes and increase the number of high‑priority issues that may warrant leadership visibility. Organizations, especially those that provide and maintain software or online services, should re-evaluate their enterprise risk programs to assess whether they are appropriately calibrated for AI-enabled cyber risks, including with respect to the risks posed to an organization’s crown jewel assets. Organizations should consider incorporating a discussion of AI-related cyber risk into briefings to the Board, relevant board committees, and senior management responsible for managing cybersecurity risks, to inform them of the evolving cybersecurity landscape. This is consistent with many cybersecurity statutory and regulatory frameworks, which prioritize the identification of risks and implementation of security measures to address those risks. In particular, NYDFS previously issued guidance indicating that AI risks should be considered under existing cybersecurity regulations. Additionally, the U.S. National Institute of Standards and Technology (“NIST”) published a preliminary draft of the Cybersecurity Framework Profile for Artificial Intelligence, which complements NIST’s Cybersecurity Framework 2.0 and provides considerations for organizations to prioritize when securing AI implementations, using AI to enhance cybersecurity defenses, and defending against adversarial uses of AI. For organizations looking to adopt agentic AI services, recent guidance from CISA (along with other government agencies) on best practices for securing agentic AI systems can be another valuable resource to assess and mitigate security risks.
Consider whether existing vulnerability management and patching processes are adequate to address AI-facilitated cyber capabilities: Frontier AI models could not only increase the volume of identified vulnerabilities but also potentially narrow the time it takes to move from identification to exploitation. Accordingly, organizations should consider whether existing procedures for vulnerability identification, remediation, and disclosure are prepared to handle the discovery of numerous vulnerabilities in quick succession by AI tools. Organizations should consider whether they have a defensible risk-based triage approach for remediation of identified vulnerabilities, including with respect to crown jewel systems and critical software provided to customers, as well as procedures for triaging vulnerabilities identified in third-party platforms and in open-source code, such as assessing risk through analysis of software bill of materials (“SBOMs”). In addition, organizations that operate bug bounty programs should consider whether to update these programs with clear rules and guardrails for AI use by security researchers, as the use of AI could drastically increase the volume of submissions while driving down their relative quality (and the benefit the organization receives from the program).
Assess your organization’s readiness to address third‑party and supply chain cyber risks: Frontier AI models’ capabilities to rapidly identify vulnerabilities across widely used components and services could also raise the stakes for third-party risk management for software vendors and service providers. Accordingly, organizations should consider how to respond to an increase in third-party vulnerability reports, including how to receive, triage, and remediate reported vulnerabilities. This could include establishing processes for escalating reports from vendors regarding critical vulnerabilities, to ensure those vulnerabilities are communicated to relevant stakeholders and remediated quickly, as well as considering options to remove or replace legacy systems and hardware that are no longer supported and at enhanced risk of exploitation. Organizations should also consider whether their contracts require vendors and service providers to disclose and remediate vulnerabilities in software products within appropriate timeframes based on the level of risk, with particular focus on critical dependencies within their crown jewel assets and supply chains, including cloud-managed services and key software providers.
Consider whether updates to SEC disclosures and related disclosure controls are warranted: The potential shift in the cyber threat landscape may prompt some publicly traded companies to reassess their processes for escalating, assessing, and disclosing cyber-related risks and potentially material incidents. While many organizations might have reviewed and revisited their disclosure processes after the SEC adopted its cybersecurity risk management and incident disclosure rules in 2023, public companies should consider whether further enhancements to these controls and procedures are necessary to address AI-related cyber risks. Additionally, public companies should review existing risk factor disclosures to ensure that AI-related cyber risks are appropriately described, taking care to avoid hypothetical risk disclosure when these risks materialize. As an example, the evolution of frontier AI models’ cyber capabilities may necessitate enhanced disclosure of risks arising from the internal deployment of AI tools, third-party dependencies, or malicious use of sophisticated AI by threat actors.
Prepare for secure adoption of enhanced AI capabilities through planning and governance: Although the cyber risks associated with deployment of AI models are not new, including the potential for internal misuse and other unintended consequences, such risks could increase as these tools are deployed more widely, including for cyber defense purposes, and as AI models become increasingly capable. In anticipation of this expansion in capabilities and deployments, organizations should coordinate discussions among key stakeholders to assess when and how these capabilities should be deployed in a manner consistent with the organization’s AI governance framework instead of leaving these questions to be resolved through siloed or isolated decision-making. Deployment of AI models without sufficiently robust processes and controls, including access controls, could cause leakage of sensitive data, damage or degrade IT systems, or facilitate malicious activity by insider threats. Before deploying frontier AI models, organizations should implement control frameworks to mitigate these risks, including an enterprise-wide AI policy, appropriate authorized-use policies, role-based trainings for employees using these systems, and procedures for restricting and monitoring access and use. These controls should also be integrated into existing governance structures and planning, such as for cybersecurity, privacy, data management, business continuity, insider threat, third-party risk management, and human resources, among others. If an organization has an AI-specific incident response playbook, it should also consider whether to account for cyber incidents involving malicious or negligent use of AI models by insiders as part of the playbook’s processes.
Update incident response procedures to account for the speed of AI-enabled cyber incidents: While frontier AI models have significant upside potential for defensive use cases, they also have the potential to enhance threat actors’ ability to quickly compromise systems beyond current AI capabilities already in use by threat actors. To address the dual-use nature of these capabilities, organizations should consider revisiting and updating their incident response and notification procedures – either in concert with a tabletop exercise (discussed below) or as a stand-alone review – to account for the speed and risk profile of AI-enabled incidents and other significant cyber events. For example, incident response plans should address (i) rapid initial assessment, (ii) accelerated containment decisions, and (iii) internal escalation processes that can be executed when the facts are still developing. Organizations might also consider whether to expand the scope of incident response or crisis management plans to cover non-malicious cyber incidents, such as the discovery of critical vulnerabilities or inadvertent data disclosure, and creating an AI-specific playbook for AI-related cyber incidents and events that includes clear roles and responsibilities for Security, IT, Legal, and other non-technical stakeholders. Organizations should also consider how AI capabilities could change the impact assessment from a cyber incident; for example, threat actors might leverage advanced AI models to rapidly identify critical or sensitive information within a large data set, which could change the evaluation of risk arising from such an incident. Organizations might also consider deploying such tools as part of their own assessments to evaluate and triage the risks from data exposure.
Develop a communications plan and evaluate potential communication needs and disclosure obligations in the event of an AI-enabled cybersecurity incident, significant vulnerability, or other cyber event: In the event of an AI-enabled cybersecurity incident, or a significant vulnerability, outage, or other form of cyber event, an organization might face the need (and, in some cases, the legal obligation) to communicate quickly and accurately with internal and external audiences about the issue. Ahead of a significant AI-enabled cyber event, organizations should consider whether they have a communications plan in place to handle such an event, including clearly defined roles and processes to decide who is authorized to speak on which topics and to ensure consistent messaging to various stakeholders. Organizations should also consider re-assessing their understanding of their disclosure obligations and how they would plan to meet them, especially in light of evolving regulatory and contractual notice obligations. For example, certain organizations may also be obligated to notify third parties in the event significant vulnerabilities are uncovered or for other AI-enabled cyber events that do not rise to the level of an “incident,” which could be useful to identify before such an issue arises.
Conduct table-top exercises using AI-related scenarios to assess and update crisis management capabilities: With the compressed timelines that will come with more sophisticated AI-enabled risks, testing the various plans and processes discussed above will be critically important to ensure organizations can react quickly and effectively when an AI-enabled cyber event occurs. Organizations should consider hosting table-top exercises (“TTX”) built around the scenario of an AI-related cyber incident or other “crisis” event with cross-functional participation from key crisis management stakeholders, such as information security, legal, communications, and other members of the organization’s senior leadership, as a best practice to assess how an organization’s crisis management or incident response plans and related procedures, as well as its key personnel, would fare in the event of an actual incident. Organizations might also consider conducting a TTX based on scenarios related to AI misuse, such as damage or degradation of IT systems caused by the use of AI on company systems by an employee or contractor, or other potential significant cyber events (such as critical outages or vulnerabilities). When conducting a TTX, organizations should consider doing so under privilege at the direction of counsel to facilitate protections for an open and honest dialogue among stakeholders to identify and address potential issues before an incident actually occurs.
Back
