The Microsoft Digital Defense Report 2026, published on 1 October 2026, describes a change that security teams can no longer ignore. Artificial intelligence is no longer used only to accelerate the work of human attackers. It now directs attack sequences autonomously: identifying the vulnerability, exploiting it, moving laterally, exfiltrating data. Humans supervise. The machine executes. This shift changes a fundamental assumption in your NIS2 and DORA programmes: that your organisation will have enough time between intrusion and detection to trigger an alert procedure.
What the report documents precisely
The Microsoft DDR 2026 analyses the threat activity recorded between January and June 2026. Two data points structure the analysis. First, AI has moved from an assistive role to a directing role in attacks: malicious groups use agentic systems to chain the steps of an intrusion automatically, from initial phishing through to data collection, without manual intervention between steps. Second, 39,000 vulnerabilities were published in the first six months of 2026 alone (source: Microsoft Digital Defense Report 2026, 1 October 2026), which puts the year on a trajectory of doubling the historical annual record.
These two data points combine. The attack surface is widening at an unprecedented pace. Tools capable of automatically exploiting newly published vulnerabilities reach malicious operators faster than before. The delay between the publication of a CVE and its active exploitation is shrinking: ENISA noted in its September 2026 Threat Landscape that 73% of organisations that suffered an intrusion had a patch available at the time of the attack that they had not yet deployed (source: ENISA Threat Landscape 2026, September 2026).
Why T0 under NIS2 becomes the breaking point
The NIS2 Directive and its French transposition require an early warning to ANSSI within 24 hours of becoming aware of a significant incident. This 24-hour period is calculated from T0, the moment the entity becomes aware of the incident.
The implicit logic of this period assumes that the intrusion unfolds over several days or weeks: the attacker gets in, moves around, establishes persistence, and then begins to exfiltrate. The organisation has a window in which to detect. If its tools work, it triggers the procedure before exfiltration is complete.
The AI-guided autonomous attack breaks this assumption. When an automated exploitation sequence can chain initial access, privilege escalation and data collection in under an hour, T0 no longer precedes exfiltration. It coincides with it, or comes after. The organisation learns of the incident at the moment the data has already left the perimeter.
This is not a hypothetical situation. Since 2025, CERT-FR has documented compromises in which the delay between initial access and the start of exfiltration fell below two hours for the most automated intrusions. The problem is not notification: it is detection.
A concrete example: the attack on a logistics provider
In August 2026, a European logistics operator falling under NIS2 suffered an intrusion exploiting a vulnerability published 11 days earlier in a component of its VPN. The sequence recorded in the logs: exploitation at 02:17, creation of a privileged service account at 02:31, connection to the HR file server at 02:44, exfiltration of 4 GB of data to an external storage service between 02:51 and 03:09. The SIEM alert triggered at 03:22, on the detection of an abnormal volume of outbound transfer.
The organisation notified ANSSI at 04:15. It was within the deadline. But at the time of notification, 4 GB of HR data, including special category information, had already left the infrastructure. GDPR notification to the data subjects became mandatory. The organisation had complied with the letter of NIS2 and of Article 33 GDPR. It had not detected the intrusion in time to stop it.
What this requires of your detection programme
The answer is not to speed up notification. It is to move detection earlier in the attack chain. The operational pitfalls of the NIS2 24-hour incident notification already include the risk of detection that comes too late. What the context of autonomous attacks adds is a new time constraint: your MTTD (Mean Time To Detect) must now be measured in minutes, not hours, for the most critical infrastructure segments.
Three adjustments are necessary.
First adjustment: behavioural coverage of privileged accounts. Automated attacks first target high-privilege credentials. A UEBA (User and Entity Behaviour Analytics) system calibrated on the normal behaviour of these accounts can detect an unusual session creation outside normal hours within the first minutes. The detection rule is not complex. It is absent from the majority of NIS2 programmes audited in 2025, according to ANSSI field teams.
Second adjustment: real-time monitoring of outbound volume. Exfiltration leaves a volumetric trace. A transfer of several gigabytes to an unregistered cloud service at 3 a.m. is detectable by a DLP (Data Loss Prevention) tool configured with dynamic thresholds. This type of detection does not require complex correlation. It requires that monitoring of outbound network activity has been enabled, which many organisations defer for reasons of cost or complexity.
Third adjustment: testing the detection chain, not only the notification procedure. Incident simulation exercises often focus on the notification procedure: who calls whom, which form, which deadline. The scenario of an autonomous attack at 2 a.m., with detection occurring after exfiltration, should be part of your exercises. The result will give you a real measure of your MTTD on the highest-risk scenarios.
What the shift towards autonomy changes for your remediation plans
The Microsoft DDR 2026 identifies a second effect of the move to autonomous AI: malicious groups can now operate at a scale that exceeds human supervision capacity. A single operator can oversee dozens of simultaneous intrusions, delegating execution to agentic systems.
For your teams, this means the volume of significant incidents you must notify could increase. NIS2 requires a notification for each significant incident. If your organisation is targeted by several simultaneous automated intrusions, your notification workload may exceed the capacity of your security team.
The answer is not to add staff. It is to have an incident management system that automates the qualification (significant or not), the documentation of the timeline and the pre-drafting of the notification, so that your teams can focus on high-stakes decisions. This is precisely what Presidio covers in its incident management module: automatic qualification against NIS2 criteria, an immutable, timestamped timeline, and a notification drafting assistant.
The regulatory requirement does not change. The threat does.
NIS2 will not be revised to account for autonomous attacks. The 24-hour deadline remains 24 hours. What has changed is that meeting this deadline now requires detection that most organisations have not yet put in place. Your ability to notify on time depends on your ability to detect before exfiltration. This is no longer a matter of procedure. It is a matter of infrastructure.