Startup discovers critical vulnerability in Java using artificial intelligence
Read more
Tecnoblog
tecnoblog.net

Startup discovers critical vulnerability in Java using artificial intelligence

A cybersecurity company has discovered a critical vulnerability in Java that could allow attackers to execute remote code and seize control of systems belonging to enterprises, banks, and government institutions. This vulnerability has been named Java4Shell (CVE-2026-114938).

The discovery was achieved through the use of artificial intelligence, specifically OpenAI's GPT-6 Cyber model. In a technical report, researcher Quan Luong also described the work of the Codex agent. The company claims that finding the vulnerability took about an hour, and developing the first functional exploit took another two hours, with AI usage costs amounting to less than $200 USD.

The problem lies within the Base64 decoder in OpenJDK. This tool is used to convert encoded information back to its original form, which is standard practice for authentication, file transfer, and inter-system communication. According to Calif, a specific sequence of 64 'A' characters followed by an equals sign can activate this vulnerability on certain installations.

The error is related to an optimization that uses specific processor instructions to speed up the decoding process. This flaw potentially can cause freezes, data leaks, and, in some cases, allow remote code execution. Calif demonstrated exploits in applications such as Keycloak, Atlassian Jira, Confluence, Jenkins, and JFrog Artifactory.

It is important to note that the vulnerability does not affect all computers running Java; it only impacts specific versions of OpenJDK running on x86-64 processors compatible with AVX-512 VBMI. Systems with ARM64 processors are not considered affected.

This finding expands on a series of vulnerabilities identified using AI. Earlier in October, the same company, Calif, participated in discovering issues in PHP that affected government servers and services.

Oracle released a security update on Tuesday (11/17) to fix this vulnerability, classifying it as high priority. The US Cybersecurity and Infrastructure Security Agency (CISA) recommended that organizations immediately install available patches. The warning also advises administrators to identify vulnerable Java versions, including those integrated into third-party applications.

Authorities in Germany and the Netherlands reported that no known successful attacks using this vulnerability were registered before Wednesday (11/18); however, this does not rule out risk, as the technical details have already become public.

Similar stories

SBI Head Warns of Systemic Failure Risks in Financial Institutions Due to AI Errors
Read more
business-standard.com

SBI Head Warns of Systemic Failure Risks in Financial Institutions Due to AI Errors

Errors related to artificial intelligence (AI) can trigger cascading failures in interconnected systems, creating serious governance challenges for financial organizations. C. S. Sethi, Chairman of the State Bank of India (SBI), stated on Thursday that to prevent systemic risks, banks must strengthen AI governance by establishing clear permissions, transaction limits, audit trails, and human oversight mechanisms, as banks increasingly adopt autonomous systems.

Sethi noted that while AI capabilities provide significant productivity gains, they also generate new governance risks. He emphasized that institutions must have a very clear understanding of what actions an autonomous system is authorized to perform, what information it can access, what decisions it can make, and when human intervention is required.

Key Responsibilities of Boards of Directors

In this context, Sethi highlighted the key responsibilities of boards of directors, including ensuring that the adoption of artificial intelligence is driven by the organization's goals rather than merely a technological impulse. According to him, technologies should solve real business problems, whether improving customer experience, strengthening risk management, or expanding access to financial services.

He added that implementation should not become an end in itself, which is particularly important in a diverse country like India. Sethi cautioned that AI should not unintentionally exclude rural borrowers, small businesses, or individuals without traditional credit histories, insisting that financial inclusion must remain a central element in the development and application of this technology.

Regarding governance frameworks, Sethi reported that SBI became the first bank in the banking system to adopt a responsible AI policy defining the principles, boundaries, and accountability for using this technology within the organization.

He advised bank directors to develop competencies in model risk assessment, data governance, cybersecurity, and algorithmic bias, while understanding the limitations of AI. Sethi compared this to annual cybersecurity drills expected from every board of directors and suggested integrating AI training into this process, emphasizing that directors must understand AI risks, even if they do not need to know how to implement it.

Sethi also mentioned the Reserve Bank of India's policy on AI, according to which ultimate responsibility for AI governance lies with the board of directors. However, he noted that AI can enhance board efficiency by identifying emerging risks, simplifying complex information, and supporting more informed decision-making, allowing directors to focus more on strategy and long-term priorities.

The Need to Maintain Human Accountability

Stressing the importance of maintaining human accountability, Sethi stated that while AI can recommend or execute actions, the final responsibility for institutional decisions must always rest with the board of directors. He pointed to the serious governance issues an institution will face if it cannot adequately explain why an AI-based process denied a loan, flagged a transaction, or initiated any action, thereby underscoring the importance of explainability, traceability, and independent verification.

He added that AI is most effective in situations where parameters can be defined, relevant data is available, and outcomes can be assessed. However, during crises, unexpected changes in the operating environment, or strategic decisions involving unfamiliar circumstances, the collective experience and independent judgment of the board remain indispensable.

Sethi concluded that the pace of AI adoption must match the institution's capacity to manage this technology, especially when decisions have significant financial, legal, ethical, or reputational consequences. He noted that corporate governance faces particular strain when the speed of technological change outpaces existing regulatory frameworks. Therefore, boards of directors must ensure that innovation proceeds without compromising the integrity of the institution and public trust.

He stressed that AI should complement, not replace, the judgments and oversight of the board of directors, enhancing their ability to fulfill their duties while keeping responsibility where it belongs. The head of the country's largest lender mentioned that his bank has used artificial intelligence and machine learning, particularly for managing fraud and operational risks. Initiatives include a patented AI and machine learning model that identifies branches that are outliers in terms of fraud management. This model assesses about 23,000 branches, classifying them into very high, high, medium, and low-risk categories. This assessment helps the bank prioritize branch audits.

Nevertheless, Sethi reported that the focus is now shifting from individual AI use cases to enterprise-wide adoption, where governance becomes particularly crucial for large institutions. He defined fairness, reliability, explainability, security, and accountability as critical elements of AI governance. He also noted that model bias is not just a theoretical problem for a bank like SBI, which serves a highly diversified client base, and that 'fairness and preventing model bias are extremely important from a governance perspective.'

OpenAI's Artificial Intelligence Discovers Critical Vulnerability in Java Capable of Granting Access to Hackers
Read more
olhardigital.com.br

OpenAI's Artificial Intelligence Discovers Critical Vulnerability in Java Capable of Granting Access to Hackers

Artificial intelligence has helped researchers discover a critical vulnerability in Java, a programming language used in systems of approximately 90% of companies on the Fortune 500 list. This vulnerability, named Java4Shell, potentially allows attackers to gain unauthorized remote access to internal systems.

The discovery was announced on Tuesday (17) by Calif, a cybersecurity startup, which utilized OpenAI's GPT-6 Cyber model designed for digital security research. According to the company, the vulnerability was found in Java's Base64 decoder, a component responsible for converting specific binary data into text.

The issue attracted attention due to its potential scale. Java is used in banks, hospitals, airlines, technology companies, and government agencies, including those related to national security. Bruce Dang, a senior researcher at Calif, stated that 'Java is the beating heart of the modern economy,' and suggested that almost all large companies would be affected by this vulnerability.

This finding also highlights a shift in the digital security sector: AI tools are accelerating the detection of software bugs, a pace that may challenge the work of system protection specialists.

In recent days, Calif has been working to coordinate the disclosure of the vulnerability with governments and technology companies. These efforts have involved countries in the Five Eyes intelligence-sharing alliance: the United States, Canada, Australia, New Zealand, and the United Kingdom. Companies such as Google and Microsoft were also notified. Oracle, which owns the Java brand and provides commercial licenses for the language, released a security update on Tuesday.

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) also published a warning recommending that organizations immediately install the fix. Google stated that after receiving information about the vulnerability, it implemented protective measures and updates in its services, and developed detection mechanisms to check for exploit usage, publishing guidelines for customers.

It is unclear whether temporary outages in the U.S. National Security Agency (NSA) network on Tuesday are related to the fix. The Times reported that the agency did not respond to a request for comment on this matter.

This discovery is the second major vulnerability identified by Calif using OpenAI models in recent weeks. Last month, the startup discovered serious flaws in PHP, a programming language used in approximately 70% of websites and applications on the internet. This incident triggered an international mobilization to patch vulnerable systems before criminals could exploit them.

At that time, government bodies, including the NSA, recommended that companies take their sites offline if they could not immediately apply the necessary patches. Despite initial concerns, the consequences turned out to be less severe than predicted, as hackers were unable to easily reproduce the attack method.

The difference is that Java is present at a broader level of digital infrastructure. In addition to standard applications, this language supports cloud computing services, identity and authentication systems, and mail platforms.

Thai Duong, CEO of Calif, noted that the vulnerability is not as simple to turn into an attack, giving companies and developers more time to apply fixes. Nevertheless, he warned that such errors can remain a threat for years; systems that do not receive updates over the next five to ten years may remain susceptible to this problem.

This case also demonstrates how AI is changing the dynamics of cybersecurity. Advanced models help researchers find code issues and develop exploit demonstrations in much less time than previously required. This can help defensive teams identify vulnerabilities sooner than criminals. On the other hand, the same tools can reduce the effort needed to create sophisticated digital attacks.

Ollie Whitehouse, CTO of the UK's National Cyber Security Centre, stated that AI is already being used to detect a growing number of software and digital service vulnerabilities. However, there is concern that the technology is also facilitating the activities of malicious actors. This week, the FBI issued a warning about hackers allegedly using autonomous AI tools to attack government agencies and companies in various countries.

Fouad Matin, head of research and cybersecurity at OpenAI, stated that Calif's research shows how AI can help specialists find serious errors in software used by millions of people. He added that collaboration between researchers and developers can make critical systems more secure.

The progress of AI in cybersecurity is reminiscent of World Chess Champion Garry Kasparov's prediction after his defeat by IBM's Deep Blue nearly 30 years ago. At that time, Kasparov suggested that the best games would be played in a combination of human and computer, a concept known as centaur, after the mythical creature combining a human and a horse. For years, players combining human abilities and computational power were considered the strongest in chess. Now, the same idea is beginning to emerge in digital security, where researchers and AI tools work together to find and exploit vulnerabilities.

Calif and other teams have already demonstrated that small groups of experts can develop attacks that previously required months of work from large, highly specialized hacker teams, in just hours or days. According to Thai Duong, the startup's CEO, AI may soon be able to develop complex attacks autonomously. 'I don't know how much time we have—maybe three months, maybe three years. But right now, the best hacker is a centaur hacker,' he told The Times.

Artificial intelligence accelerates cyberattacks and increases the number of incidents in companies
Read more
olhardigital.com.br

Artificial intelligence accelerates cyberattacks and increases the number of incidents in companies

Progress in artificial intelligence (AI) coincides with a rise in cyber incidents, according to data from the insurance company Marsh. This technology accelerates the pace and scale of attacks, creating new problems for both companies and insurance organizations.

In the last two weeks of October, Marsh clients reported 105 cyber incidents. This volume is more than double the 48 cases registered in the last two weeks of May.

In an interview with Reuters this week, Meredith Schnur, who is responsible for Marsh Risk's cyber risk practice in the US and Canada, noted signs that attackers are using AI tools to find vulnerabilities in systems.

"There are signs that attackers are using AI-based and agent tools to identify vulnerabilities. So I would say that we can attribute a significant number of these incidents to AI," stated the head.

Among the exploited weaknesses are unpatched software, open systems, and weak identification mechanisms. Artificial intelligence allows criminals to find and exploit these vulnerabilities much faster.

The popularization of such tools also facilitates the expansion of criminal activity at lower costs, increasing the potential number of targets. According to Schnur, this trend, combined with the slow response of organizations to attacks and incidents, explains a significant part of the recent increase in registrations.

"AI itself is not a threat; it is an amplification of risk in all areas," emphasized the head.

The increased frequency of attacks should also stimulate the digital risk insurance market. According to GlobalData forecasts, cyber insurance premiums—the amounts companies pay for policies—are expected to grow globally faster over the next two years, following the increased frequency of attacks.

The consulting firm predicts that the global market will grow by 6.9% in 2026, reaching USD 16.6 billion (approximately BRL 94.62 billion). In 2027, growth is expected to reach 10%, increasing the volume to USD 18.3 billion (approximately BRL 104.31 billion).

Popular