This is A MoFo Privacy Minute, where we provide our clients short answers to their most asked questions.
Question: Threat actors are using agentic AI to launch attacks that can reach across entire corporate networks in minutes. If an attack potentially exposes most or all of an organization’s systems and data, how can it determine what actions to take?
Answer: The era of AI-enabled attacks has arrived, and with it complex questions about whether traditional incident response practices and legal frameworks will hold. While existing statutory, contractual, and ethical breach notification considerations continue to apply, the task of determining what information and systems may have been affected, and how to communicate about an incident has become significantly more challenging. Organizations should evaluate whether their current incident response processes are ready for a world in which the “blast radius” is exponentially larger, the legal questions presented even more complex, and business disruption becomes commonplace.
The Landscape Has Fundamentally Changed.
Governments and regulators are sounding the alarm about the recent tsunami of AI-enabled incidents. The UK’s National Cyber Security Centre (NCSC)[1] has warned that, by 2028, the UK is highly likely to face an increase in the scale and speed of sophisticated cyberattacks, and that AI-powered malware is highly likely to be used in a ransomware attack on the UK’s critical national infrastructure.
Threat actors now need only to provide high-level instructions that are translated by AI into sequences of rapid-fire technical commands that, with limited human intervention, relentlessly pursue a particular goal until success or failure are objectively reached. That goal could be obtaining credentials, simply experimenting with a new AI tool to see if it’s effective, or bringing an entire system offline.
For the past 20 years, incident response has often involved taking immediate steps to contain and secure systems, then working to reconstruct an attacker’s activities, and carefully analyzing affected data. Organizations must now grapple with how and whether that will continue to be feasible where incidents have impacts across hundreds of interconnected systems and vastly different data sets.
Notification Frameworks May Break.
- How is an organization to define “access” when an AI agent may touch everything? Existing breach notification frameworks were not designed to address autonomous intrusions. Under various global notification frameworks, unauthorized access to personal information may constitute a notifiable breach, but does an AI agent’s ability to reach a system, test permissions, or catalog databases mean that access has occurred? How does an organization assess whether that access may cause an individual harm? Organizations should begin considering how they will answer these questions using real world examples.
- If an attack potentially reaches everything, must organizations notify everyone? Organizations should begin considering how they will meet their legal obligations where the scope of an incident is potentially immense but what actually occurred is unclear.
Many breach notification frameworks already allow organizations to make risk-based assessments. In addition, current reforms such as those proposed by the European Commission, and supported by the European Council, would raise the EU GDPR’s threshold for required notices to data protection authorities to breaches that are likely to result in a high risk of harm.[2]
Over time, we may see regulators become more receptive to organizations taking risk-based holistic approaches rather than imposing what will quickly become an impossible task as AI-enabled attacks continue. The practical realities of cost, speed, and precision of notice will continue to be front and center as organizations consider how to respond in a defensible way.
What Should Organizations Be Doing Now?
Organizations can take steps now to prepare to respond to AI-enabled incidents so they are prepared to face difficult decisions, often under significant time pressure.
- Rethink how data impacts are assessed. Develop approaches that distinguish an attacker’s potential access from actual validated data access, use, alteration, collection, or exfiltration. Given the volume of logs generated by an AI-enabled attack, a comprehensive forensic review may be impracticable or even impossible. Consider how your organization may be able to prioritize or assess the potential risks to systems or categories of affected information.
- Evaluate what constitutes reasonable security measures. Agentic attacks can produce substantial volumes of detectable activity, but the speed of the attack may make traditional response tactics insufficient. As a result, what regulators, courts, and contracting parties regard as reasonable security is likely to evolve. Telemetry that flags individual events in isolation may no longer be adequate. Instead, consider whether security monitoring tools can examine patterns and sequences of activity across systems, and detect and immediately disrupt an attack in progress. If automated disruption becomes an expected feature of reasonable security, organizations should also consider whether temporary service disruption will become more common and how they will navigate this challenge.
- Prepare to make difficult containment decisions. Organizations may need to respond to agentic intrusions by automatically suspending accounts, revoking credentials, isolating systems, or temporarily disabling services. These measures may prevent further compromise but may also disrupt legitimate operations. Legal, security, and business teams should consider in advance when such measures may be justified, who can authorize them, and whom decision-makers must consult before acting. This will be very different for each organization based on its service. Considerations regarding having bank account access delayed will look very different from incidents where connected devices put health and safety risks center stage. Start planning now for what will make sense for your organization so you are ready to act quickly when an attack occurs. Consider potentially affected customers, critical services, and contractual commitments now as you prepare.
- Revisit notification strategies and decision-making processes. Incident response tabletop exercises should account for situations where the organization is still working to establish the scope of affected data and systems, and quick decisions need to be made. Organizations should practice how to navigate incident response against a backdrop of large-scale forensic investigations, system disruptions, and rebuild activities, and pressing regulatory reporting and stakeholder communications.
Looking Ahead
Agentic AI has fundamentally changed how organizations approach incident response in reality. It is important to begin preparing now so you can act decisively for your organization, even when working with incomplete information and immense data sets.
[1] See NCSC, Artificial intelligence and cyber intrusion, November 3, 2026.
[2] See European Council, Digital rulebook: Council agrees position to simplify data protection, privacy and cyber reporting rules, October 21, 2026.