OpenAI and Hugging Face are investigating an AI security incident involving an AI agent that compromised infrastructure while models were being evaluated for advanced cyber capabilities. The incident was detected and contained after the models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure.

Hugging Face disclosed the incident last week, while OpenAI has now shared preliminary findings from its investigation. According to OpenAI, the activity involved a combination of models, including GPT 5.6 Sol and a more capable pre-release model, with reduced cyber refusals for evaluation purposes. The models were being tested against a benchmark designed to measure advanced cyber capabilities.

OpenAI described the incident as unprecedented and said it is continuing its investigation alongside Hugging Face. The company said the findings are being shared to help defenders understand what occurred and assess the capabilities of increasingly advanced AI models.

OpenAI and Hugging Face Investigate AI Security Incident

The incident took place during an internal evaluation designed to test models against advanced exploitation scenarios involving complex attack paths. OpenAI said the evaluation was conducted without production classifiers that normally prevent models from pursuing high-risk cyber activity.

The testing environment was designed to be highly isolated, with network access limited to package installation through an internally hosted third-party software system that functions as a proxy and cache for package registries.

Despite these restrictions, the models spent substantial inference compute attempting to obtain open Internet access. During this process, they identified and exploited a zero-day vulnerability in the package registry cache proxy. OpenAI said the vulnerability has since been responsibly disclosed to the vendor.

The models then carried out privilege escalation and lateral movement within the research testing environment until they reached a node with Internet access.

[caption id="attachment_113283" align="aligncenter" width="600"] Source: OpenAI[/caption]

Models Chained Vulnerabilities Across Infrastructure

After obtaining Internet access, the models inferred that Hugging Face could host models, datasets and solutions related to the ExploitGym evaluation. The models then searched for ways to obtain secret information that could help them bypass the evaluation.

In one example described by OpenAI, the models chained multiple attack vectors, including stolen credentials and zero-day vulnerabilities, to identify a remote code execution path on Hugging Face servers.

OpenAI's security team detected the anomalous activity internally. Hugging Face's security team and agents also detected and stopped the activity on its infrastructure. The company had already begun containment and forensic reconstruction using its own open-source models when the two organizations connected.

Both companies are continuing to investigate the incident and its impact.

OpenAI Strengthens Security Controls After Incident

As part of its response, OpenAI said it is implementing stricter infrastructure configuration controls while identified vulnerabilities are patched. The company acknowledged that these measures could come at the cost of research velocity and said its Safety and Security Committee is being briefed on the controls and their impact.

OpenAI is also working with Hugging Face on the forensic investigation and has responsibly disclosed the identified zero-day vulnerability in the internally hosted third-party software.

The company has also brought Hugging Face into its trusted access program and is supporting its teams in using AI model capabilities to strengthen defenses.

OpenAI said it is improving protections around future training and evaluations, including stronger safeguards for model alignment, cybersecurity and monitoring during internal testing. The company noted that deployment safeguards were intentionally disabled during this evaluation because the goal was to measure cyber vulnerabilities.

AI Cyber Capabilities Raise New Security Concerns

OpenAI said the incident demonstrates the need for AI security and safety measures to keep pace with rapidly advancing model capabilities. The company is strengthening containment, monitoring, access controls and evaluation practices used during model development.

The incident also highlights how advanced models can potentially discover and exploit novel attack paths in real-world systems without access to source code. OpenAI said increasingly capable models should also be used defensively to help security teams identify weaknesses, understand vulnerability chains and accelerate remediation.

Hugging Face CEO Clem Delangue said the incident demonstrates the importance of collaboration in addressing AI safety and security challenges. Both organizations said they will continue investigating the incident and share additional findings and best practices as the work progresses.

A Hutt City Council phishing attack reported in March 2026 has led to the exposure of sensitive information belonging to hundreds of individuals, prompting the council to strengthen its cybersecurity measures and notify affected residents. According to officials, the Hutt City Council phishing attack resulted in unauthorized access to several email accounts. Initial investigations confirmed that identity information of five individuals was compromised, while financial details of up to 732 people may have been exposed through email correspondence.

Details of the Hutt City Council Phishing Attack

The Hutt City Council phishing attack involved malicious emails designed to trick users into revealing login credentials or granting access to internal systems. Once access was obtained, attackers were able to view email communications containing personal and financial data. Council authorities stated that while only a small number of individuals had confirmed identity data compromised, a significantly larger group may have had information exposed indirectly through email threads. All individuals impacted by the Hutt City Council phishing attack have been contacted directly and provided with guidance on steps to secure their information and reduce potential risks.

Immediate Response and Containment Measures

Following the Hutt City Council phishing attack, the organization initiated a rapid response to contain the breach and prevent further unauthorized access. This included securing affected accounts, reviewing system access logs, and strengthening internal security settings. Chief Executive Jo Miller confirmed that the incident has been reported to the Office of the Privacy Commissioner. She acknowledged the seriousness of the breach and its impact on the community. “We are sorry this has occurred and acknowledge the concern it may have caused. It’s a reminder to handle data with sufficient care,” Miller said, adding that additional safeguards have been implemented to prevent similar incidents. The council has also accelerated its cybersecurity improvement program in response to the Hutt City Council phishing attack, focusing on enhanced monitoring and faster incident detection.

Strengthening Systems and Security Controls

In response to the Hutt City Council phishing attack, several measures have been implemented to improve system resilience. These include:

• Enhanced email security settings
• Increased monitoring of account activity
• Additional staff training to identify phishing attempts
• Strengthened access controls

The council stated that these improvements are part of a broader effort to reduce the risk of similar incidents in the future.

Growing Threat of Phishing Attacks

The Hutt City Council phishing attack reflects a wider trend of increasingly sophisticated cyber threats. Authorities noted that cybercriminals are using advanced tools, including artificial intelligence, to automate phishing campaigns, making them more convincing and harder to detect. These evolving tactics allow attackers to scale operations quickly, adapt to security measures, and target organizations more effectively. As a result, early detection and rapid response have become critical components of cybersecurity strategies. The incident serves as a reminder for both organizations and individuals to remain cautious when handling emails and sharing sensitive information.

Advisory for Affected Individuals

Following the Hutt City Council phishing attack, affected individuals have been advised to:

• Monitor bank and financial statements closely
• Be alert to suspicious emails or communications
• Update passwords and enable additional security measures where possible

The council has also encouraged prompt reporting of any unusual activity to minimize potential harm.

Ongoing Review and Community Assurance

The Hutt City Council phishing attack is currently under review as part of ongoing efforts to strengthen data protection practices. Officials have emphasized their commitment to safeguarding personal information and improving system security. While the incident has caused concern, the council maintains that steps have been taken to contain the breach and reduce the likelihood of future attacks. Additional safeguards and monitoring systems are now in place as part of the response to the Hutt City Council phishing attack. Authorities continue to work with relevant agencies to ensure compliance and maintain transparency as investigations progress.

A newly identified cyber campaign involving JanaWare ransomware is targeting users in Turkey, with researchers linking the activity to a customized version of the Adwind Remote Access Trojan (RAT). The findings come from an analysis by researchers at Acronis’ Threat Research Unit (TRU), who identified the threat cluster during an investigation into suspicious Java-based malware samples. According to the researchers, the JanaWare ransomware operation appears to have been active since at least 2020. Evidence from malware samples and infrastructure indicates that the campaign has continued into late 2025, suggesting sustained activity with limited visibility. The attack relies on a modified Adwind RAT that includes polymorphic capabilities. This allows the malware to change its structure across infections, making detection more difficult. Combined with code obfuscation, these techniques have likely contributed to the campaign remaining relatively unnoticed. Unlike large ransomware groups that focus on high-value enterprise targets, JanaWare ransomware appears to follow a different strategy. Observed ransom demands range between $200 and $400, pointing to a model that prioritizes volume over large individual payouts.

Phishing Identified as Primary Infection Vector

The JanaWare ransomware campaign primarily spreads through phishing emails. Victims are lured into clicking malicious links, which lead to the download of a Java archive file. In many observed cases, the payload is hosted on cloud storage platforms. Telemetry data reviewed by researchers shows a consistent attack chain. A phishing email is opened in Microsoft Outlook, followed by a browser session that downloads the malicious file. The file is then executed using Java, triggering the infection. [caption id="attachment_111347" align="aligncenter" width="761"] Image Source: Acronis’ Threat Research Unit (TRU)[/caption] User reports on public cybersecurity forums also describe similar incidents, supporting the assessment that phishing is the main entry point.

Geofencing Restricts Janaware Ransomware Attacks to Turkey

A key feature of the JanaWare ransomware is its use of geofencing. The malware is designed to execute only on systems that meet specific regional criteria linked to Turkey. It checks system language, locale settings, and external IP geolocation before proceeding. If the system does not match Turkish parameters, the malicious activity is halted. Researchers note that this approach likely serves both operational and defensive purposes. It allows attackers to focus on a specific region while reducing exposure to global security monitoring and automated analysis systems.

Obfuscation and Polymorphism Hinder Detection

The JanaWare ransomware incorporates multiple techniques to evade detection. Researchers identified the use of known obfuscation tools such as Stringer and Allatori, alongside custom methods that complicate analysis. The malware also includes a self-modifying component that alters its file structure during deployment. By adding random data to its Java archive, each instance generates a unique file hash, limiting the effectiveness of signature-based detection. In addition, the malware contains embedded configuration parameters that control its behavior. These include command-and-control server details, communication ports, and authentication values used during initial connections.

Security Controls Disabled Before Encryption Stage

Before encrypting files, the malware attempts to weaken system defenses. It executes commands to disable Microsoft Defender, suppress security alerts, and remove recovery mechanisms such as Volume Shadow Copies. It also interferes with Windows Update and scans for installed antivirus software. These steps reduce the likelihood of detection or recovery once the ransomware payload is activated. The encryption process is carried out by a secondary module delivered after the initial compromise. This module uses AES encryption and communicates with command-and-control infrastructure over the Tor network.

Turkish-Language Ransom Notes Signal Targeted Approach

After encryption, the malware drops ransom notes across affected systems. These notes are written in Turkish and instruct victims to contact the attackers through encrypted communication channels such as qTox or Tor-based websites. Researchers say the consistent use of Turkish-language content, combined with geofencing, indicates a deliberate focus on users in Turkey rather than a broad, global campaign. The JanaWare ransomware campaign highlights how targeted, lower-profile operations can persist over long periods without drawing significant attention. By focusing on home users and small businesses, and keeping ransom demands relatively low, the attackers appear to maintain a steady but less visible operation. Researchers caution that such localized campaigns may continue to operate alongside larger ransomware groups, adding another layer to the evolving threat landscape.