Cybersecurity is moving from detecting attacks towards containing and repairing them at machine speed.
Researchers have documented what they describe as the first ransomware operation executed from beginning to end by an autonomous AI agent. Municipalities continue to struggle with limited security budgets and fragmented infrastructure, while Israeli startup Remedio is betting that automated remediation will become more valuable than generating additional alerts.
Japan is urging local governments to strengthen their defences, and agricultural businesses are discovering that connected machinery, cloud platforms and data-driven operations have transformed farms into critical digital infrastructure.
These developments point towards a new security reality: organisations cannot assume that every attacker will move at human speed, and they cannot protect increasingly connected operations through monitoring alone.
JadePuffer reportedly executed ransomware without human oversight
A ransomware operation known as JadePuffer has been described as the first documented attack in which an AI agent completed the entire intrusion chain without a human directing its individual actions.
According to Cybersecurity Ventures, the agent conducted reconnaissance, collected credentials, moved between systems, adapted when commands failed and ultimately encrypted production data.
The initial access reportedly came through CVE-2025-3248, a critical remote-code-execution vulnerability affecting Langflow, an open-source platform used to build applications powered by large language models.
From that foothold, the agent searched environment variables for sensitive information and identified credentials connected to cloud infrastructure, AI providers and internal services. It then moved towards a production environment running Alibaba’s Nacos configuration and service-discovery software.
That second stage exploited CVE-2021-29441, an authentication-bypass vulnerability affecting older Nacos deployments, together with a publicly documented default signing key.
The important development was not the discovery of an entirely new exploitation technique. Both vulnerabilities were already known, and the wider attack depended heavily on familiar weaknesses:
- An internet-facing service that had not been patched
- Default credentials and cryptographic keys
- Excessive access to secrets stored in environment variables
- Weak separation between development and production systems
- Insufficient monitoring of automated activity
The AI agent’s contribution was its ability to combine those weaknesses into a complete operation.
Researchers observed it correcting failed commands, changing its method when an authentication step did not work and adapting its parsing after receiving an unexpected response. One recovery cycle reportedly took approximately 42 seconds.
Once it reached the Nacos database, the agent encrypted 1,342 configuration records and deleted the original tables. The encryption key was apparently generated but never stored or transmitted, making recovery impossible even if the victim had attempted to pay.
The ransom note also reportedly used a Bitcoin address resembling an example address from technical documentation. This may indicate that the system reproduced a familiar placeholder without an attacker reviewing the output carefully.
These anomalies raise an important distinction. The incident does not necessarily prove that an AI independently decided to launch a ransomware campaign. A human almost certainly established the objective, supplied access or deployed the agent.
What appears to have changed is the level of operational supervision. Once activated, the agent could reportedly select and adjust technical actions without waiting for step-by-step instructions.
That dramatically compresses the defensive window. Processes designed to identify a human attacker over several hours may be ineffective when an agent can move from exploitation to destructive action in minutes.
Municipalities remain exposed by fragmented systems and limited resources
Municipal governments are becoming increasingly attractive targets because they combine valuable data, essential public services and frequently outdated technology.
Research covered by Tech Xplore highlights the difficulty municipalities face when trying to improve cybersecurity across diverse administrative systems and public infrastructure.
Local authorities may operate networks supporting:
- Tax collection and public payments
- Civil records and permit applications
- Schools, libraries and social services
- Emergency communications
- Water and waste management
- Public transport
- Local healthcare services
- Building, traffic and environmental systems
Many of these platforms were introduced by different departments at different times. They may rely on separate suppliers, incompatible security controls and software that cannot be replaced without interrupting services.
Smaller municipalities face a particularly difficult problem. They are expected to meet many of the same security requirements as national institutions or large companies but may lack dedicated security teams, continuous monitoring and specialist incident-response capabilities.
Attackers understand this imbalance.
A compromised municipal account may expose personal information or provide access to payment systems. Ransomware can interrupt services that residents cannot simply obtain from another provider. A city cannot easily suspend water, emergency or administrative operations while rebuilding its network.
Local-government cybersecurity consequently needs to be treated as a shared public-sector responsibility rather than a collection of isolated IT projects.
National and regional authorities can help by providing shared security operations centres, centralised procurement, vulnerability-management services, incident-response retainers and standardised reporting mechanisms.
Municipalities also need accurate inventories of their assets and suppliers. An authority cannot protect a system it does not know exists, especially when external contractors maintain applications or remote-access services on its behalf.
HIPTHER previously examined the wider public-sector and infrastructure challenge in its Cybersecurity Roundup covering emerging threats, institutional partnerships and critical-infrastructure defence.
Remedio argues that detecting vulnerabilities is no longer enough
Israeli cybersecurity startup Remedio is positioning automated remediation as the next major stage of enterprise defence.
As Ynet reports, most security platforms concentrate on discovering vulnerabilities and notifying administrators. Remedio instead aims to automate the process of fixing problems safely and at scale.
The distinction matters because many organisations do not suffer from a lack of alerts.
Security teams routinely receive vulnerability reports from endpoint tools, cloud platforms, penetration tests, asset scanners and threat-intelligence services. The challenge is determining which findings present genuine risk and resolving them without disrupting business operations.
A single enterprise may have thousands of known vulnerabilities. Some affect systems that are not externally accessible, while others sit directly on likely attack paths. Treating every finding as equally urgent creates an impossible workload.
The JadePuffer incident illustrates why remediation speed is becoming more important. The attack reportedly relied on vulnerabilities that had already been disclosed and patched. The defensive failure was not an absence of information; it was the continued presence of exploitable systems.
Automated remediation could help organisations:
- Prioritise vulnerabilities according to exposure and business impact
- Test patches before wider deployment
- Apply configuration changes consistently
- Rotate compromised credentials
- Isolate vulnerable services
- Verify that corrective action succeeded
- Reverse a change when it causes operational problems
Automation nevertheless introduces its own risk. A remediation platform may have extensive administrative privileges across endpoints, cloud services and production environments. If compromised or incorrectly configured, it could create disruption on a scale comparable to the attack it was intended to prevent.
Safe remediation therefore requires staged deployment, approval thresholds, immutable logs and reliable rollback mechanisms. High-risk changes should still involve human authorisation, especially when they concern safety-critical or operational systems.
The future of vulnerability management will not be defined by how many findings a dashboard can display. It will be defined by how quickly organisations can remove meaningful attack paths without damaging the services they are protecting.
Japan asks local governments to strengthen cybersecurity
Japan’s central government has asked local authorities to improve their cybersecurity following a series of incidents affecting public organisations.
The request, reported by Nippon.com, reflects growing concern about the resilience of municipal networks and the sensitive information stored within them.
Local governments manage extensive personal and administrative data. They also depend on digital platforms to deliver services to residents, coordinate emergency responses and communicate with national systems.
A successful intrusion can consequently affect more than the confidentiality of individual records. It may interrupt public services, delay payments, undermine emergency operations or allow attackers to move between connected government networks.
National guidance can establish a common baseline, but instructions alone will not correct structural weaknesses.
Authorities need funding, technical assistance and access to trained personnel. Security standards should also account for the differences between major cities and small municipalities that may have only a handful of general IT employees.
Priority measures should include:
- Multi-factor authentication for administrative and remote access
- Separation of critical services from ordinary office networks
- Rapid installation of security updates
- Offline and immutable backups
- Centralised logging and continuous monitoring
- Restrictions on contractor access
- Regular incident-response exercises
- Mandatory reporting of significant attacks
Supply-chain oversight is especially important. Municipalities frequently depend on shared software providers, managed-service companies and outsourced data centres. A vulnerability affecting one widely used supplier could expose several authorities simultaneously.
Japan’s warning therefore reinforces the case for coordinated government cybersecurity. Local autonomy must not become local isolation when threats cross administrative boundaries instantly.
Connected agriculture becomes critical digital infrastructure
Agriculture is increasingly dependent on software, connectivity and data, making cybersecurity an operational and food-security concern.
An analysis published by Engineering News argues that agricultural companies must treat digital security as seriously as physical security and biological risk.
Modern farms and food producers may use:
- GPS-guided machinery
- Connected irrigation systems
- Automated feeding and milking equipment
- Soil and weather sensors
- Drones and satellite imagery
- Cloud-based farm-management platforms
- Digital payment and procurement systems
- Automated packing and cold-storage facilities
These technologies can improve yields, reduce waste and help businesses respond to changing weather conditions. They also connect physical operations to software that may be vulnerable to ransomware, credential theft or remote manipulation.
An attacker does not need to destroy machinery to cause substantial damage. Disabling irrigation during a critical growth period, interrupting refrigeration, changing feeding schedules or preventing access to operational data can create immediate financial losses.
Agriculture also presents several security challenges that differ from those of a conventional office.
Equipment may remain in use for many years and depend on proprietary software that receives limited updates. Connectivity may be inconsistent in rural areas, making central monitoring difficult. Seasonal workers and external technicians may require temporary access, while suppliers remotely maintain specialised systems.
Smaller agricultural businesses may also assume that cybercriminals are interested only in large corporations. In reality, attackers often target organisations that cannot tolerate downtime and may therefore feel pressured to pay quickly.
A practical agricultural security programme should begin with identifying which systems could interrupt production. Critical operational technology should be separated from email, office and guest networks, while suppliers should receive only the access necessary for their work.
Backups must include operational configurations and not merely financial documents. A farm may recover its accounting files yet remain unable to restart an automated production line because equipment settings were not preserved.
The growing exposure of agricultural technology resembles the risks long associated with energy and water infrastructure. HIPTHER has previously explored this convergence through its report on evolving cybersecurity threats to hydropower dams.
The bigger picture: cybersecurity must move from alerts to resilience
The five developments reflect the same structural change.
Attackers are gaining tools that can discover weaknesses, adapt techniques and move through networks with less direct human involvement. Defenders, meanwhile, remain constrained by fragmented systems, incomplete asset inventories and remediation processes that often move much more slowly than the threats.
Municipalities, farms and public-service organisations are particularly exposed because their digital systems control outcomes in the physical world. A compromised network can interrupt water, transport, food production or access to government services.
The response cannot be another layer of alerts.
Organisations need to reduce the number of exploitable paths before an attack begins. That means patching exposed systems, removing default credentials, separating critical networks, limiting administrative privileges and ensuring that recovery procedures work under real operating conditions.
Automation will become essential because human teams cannot manually process every vulnerability or suspicious event. However, defensive agents require the same safeguards expected of any powerful autonomous system: limited permissions, observable actions, approval gates and reliable shutdown mechanisms.
The central question is no longer whether AI will participate in cyberattacks. It already appears capable of completing substantial parts of the intrusion lifecycle.
The more important question is whether defenders can build systems that contain attacks at comparable speed—without surrendering control of their own infrastructure in the process.
