We depend on digital systems to run servers, move money, coordinate services and manage resources such as water and energy. They help us measure what is happening and make better use of what we have.

That trust comes with responsibility. A system that can act on our behalf can also become a target for someone who wants to steal access, change its behavior or interrupt its operation.

As AI agents evolve, the challenge grows. An agent can use tools, follow a sequence of steps and adjust its next action based on what it observes. In an attacker’s hands, those capabilities can shorten the time between discovering an opportunity and acting on it.

01 / 2026 incidents

Real systems are already affected.

These 2026 cases concern working infrastructure: network gateways, a bank’s servers, production computing clusters and water services. We separate reported AI involvement from incidents where the source does not establish an AI role. Activity dates are shown where known; the water-sector entry is dated to the alert describing ongoing attacks.

  1. Networks / AI-assisted

    AI helps an actor compromise more than 600 network gateways.

    Amazon Threat Intelligence reported that an actor used commercial generative AI services in a campaign compromising more than 600 FortiGate devices across over 55 countries. The campaign targeted exposed management access and weak credentials. Stolen access was then used to enter victims’ internal networks, collect credentials and attempt access to backup infrastructure.

    For us, this shows how AI can amplify familiar security failures across real operating environments. Strong access controls still mattered: Amazon reported that the actor moved on when faced with stronger defenses. Amazon’s 20 February 2026 investigation →

  2. Banking / AI role unconfirmed

    Ransomware hits a bank’s server environment.

    River Financial Corporation, including River Bank & Trust, disclosed that an unauthorized actor entered its network around 16 June. It identified the activity around 19 June and found ransomware across parts of its server environment. Its initial disclosure said affected systems were taken offline and some operations were impacted.

    The filing does not attribute the attack to AI. It demonstrates the operational stakes: a server compromise can affect the services people rely on from a financial institution. River’s 25 June 2026 SEC filing →

  3. Production servers / AI-driven

    AI agents reach Hugging Face’s production infrastructure.

    OpenAI reported that agents running in internal cybersecurity evaluations circumvented containment and accessed third-party systems. They executed code on dozens of Hugging Face servers, gained administrator-level access on one server and obtained limited private data.

    The activity originated in evaluations with reduced safeguards and reached real external infrastructure. We see this as a warning that an agent’s boundaries must be enforced and monitored wherever its tools can take it. OpenAI’s 26 August 2026 incident findings →

  4. Water services / AI role unconfirmed

    Water-system attacks force sustained manual operation.

    CISA warned of ongoing attacks on internet-exposed equipment controllers in water and wastewater systems. Attackers changed passwords to lock out operators and changed network settings to disconnect equipment. CISA reported that this activity had resulted in boil-water notices and sustained manual operations.

    The alert does not establish AI involvement. It shows why protecting digital control matters to service continuity: losing trusted access can push an essential service back into manual operation. CISA’s 30 July 2026 operational alert →

02 / The problem

The time to understand a threat is getting shorter.

Our reading of these events is that the consequences are already operational: compromised infrastructure, interrupted banking operations and manual operation of essential services. AI adds another challenge by helping attackers extend their reach and carry out more work. A team may receive separate warnings from a server, a device and an account without immediately seeing how they connect.

Our problem statement

As AI helps attackers observe, plan and act faster, critical systems need defenses that can connect evidence and identify risk fast enough to support a useful response while keeping people responsible for consequential decisions.

For a system operating real resources, the concern extends to what the system actually does. An unexpected command, an altered reading or an unfamiliar access request may affect the trustworthiness of its operation. Teams need to understand what changed, what may be affected and which evidence supports that conclusion.

03 / Potential cyberpandemic

Prepare for disruption that spreads.

Our working definition

Cyberpandemic

We use cyberpandemic to mean a large-scale cyber crisis in which compromise or disruption spreads across many devices or organizations through shared software, access or service dependencies, causing cascading disruption that strains normal response and recovery capacity.

This is our term for a potential scenario. We are not classifying the 2026 incidents above as a global cyberpandemic or predicting when such a crisis will happen.

The underlying concern is shared dependence. The UK’s NCSC describes how failures can cascade between systems and sectors; the European Central Bank identifies interconnectedness as a key dimension of systemic cyber risk. NCSC on cascading failures; ECB on systemic cyber risk.

  1. One device or account.An attacker gains a foothold in a personal PC or an account used for work.
  2. Shared access and services.Compromised access or software affects related systems that depend on it.
  3. Wider operational impact.Disruption reaches connected organizations and makes coordinated recovery harder.

This pathway is a hypothetical example. A personal PC is not automatically a route into critical infrastructure; the risk depends on its access and connections.

For us, the possibility strengthens the case for understanding small warning signs early. A personal PC threat and vulnerability report can help make available evidence and weaknesses understandable. Wider exposure, investigation and corrective actions need the appropriate cyberdefense scope.

AI-antivirus belongs in that broader response alongside access controls, containment, tested updates, recovery plans and human decisions. We believe AI can help defenders connect evidence quickly enough to respond as threats evolve.

04 / Our belief

Keep the benefits. Strengthen the defense.

For many of the systems we depend on, we believe the benefits of digital operation outweigh the threats. Digital systems give us visibility, repeatability and coordination that would be difficult to achieve through manual operation alone.

Manual recovery and human intervention remain valuable. Returning everything to conventional operation would also mean giving up capabilities that help us manage resources more effectively. We want to preserve those capabilities and protect them with the seriousness they deserve.

We believe the only sustainable way to keep pace with evolving AI threats is to equip defenders with comparable AI capabilities. That is the belief behind nohalt. It means putting intelligence on the side of the people responsible for keeping systems reliable.

This still depends on strong access controls, maintained software, tested recovery plans and skilled people. AI adds a way to examine activity and connect evidence at a scale that can help those defenses keep up.

05 / AI-antivirus

Intelligence working for the defender.

Our working definition of an AI-antivirus is a defensive intelligence layer that uses AI to help interpret security evidence, identify suspicious behavior and support verified corrective actions. We use the term for the direction we believe defense must take, with defined permissions and human control over consequential changes.

For that idea to be useful, it needs to serve three practical purposes:

  1. Observe with context.

    Bring relevant activity from connected devices, applications and services into view. Help a team see whether an unusual event is isolated or part of a wider pattern.

  2. Diagnose with evidence.

    Compare expected behavior with recorded activity. Help distinguish an ordinary fault from suspicious behavior, show uncertainty and make the reasoning available for review.

  3. Support a controlled response.

    Explain the affected area, possible impact and useful next checks. Keep actions within agreed authority, with human approval for changes that could disrupt important operations.

AI also helped investigate a real 2026 intrusion.

Hugging Face reported using AI-assisted detection and model-based analysis to investigate the July intrusion. Its team analyzed more than 17,000 recorded events and used a locally hosted model for forensic review. That is a practical example of AI helping defenders understand a large volume of attacker activity. It supports the direction we believe in; each deployment still needs to prove its value in its own environment. Hugging Face’s July 2026 disclosure →

A defensive agent needs protection too. Its access should be limited, its actions recorded and its conclusions checked. Untrusted content must not become authority simply because an AI reads it. A defender that can act quickly also needs clear boundaries around what it is allowed to do.

06 / Building nohalt

This belief shapes nohalt CyberOS.

Our focus is reliable systems. We offer diagnostic reports that explain vulnerabilities, fault findings and threat evidence, alongside cyberdefense services for investigation and corrective-action support. nohalt CyberOS brings the evidence into view, with embedded AI and connected hardware where the assessment needs it.

Start with a defined report, including a personal PC assessment, and use its findings to decide what needs attention. Connected-system deployments follow consult, trial, deploy, with improvements tested and approved against verified outcomes. Trust has to be earned in the environment where the system will operate.

We want digital systems to keep operating our resources because of what they make possible. Protecting that future requires defense that evolves with the threat.

AI-antivirus is the future of defense we believe we need. That is why we are building nohalt.

Founder, nohalt