Cyber Brief: Oracle flaw, AI poisoning, phishing tactics

Today's stories highlight the importance of understanding how attackers are evolving their techniques beyond simple exploitation. A critical Oracle vulnerability shows how threat actors now target legitimate functionality rather than just software flaws, while new research reveals how AI models can be poisoned through web pages and email summaries. Meanwhile, CISA's red team assessments demonstrate the operational difference between having security tools and actually detecting attacks, and a sophisticated phishing platform shows how real-time session hijacking is becoming industrialised.

Critical Oracle vulnerability exploited despite patching efforts

The Register reports that a perfect-10 CVSS vulnerability in Oracle Fusion Middleware is being actively exploited, with CISA adding it to its Known Exploited Vulnerabilities catalogue and mandating US federal agencies patch within three days. The flaw, CVE-2026-60004, was disclosed in January and allows unauthenticated remote code execution. What makes this particularly concerning is that organisations could have applied all 1,449 Oracle patches released this year and still been vulnerable, because attackers are now exploiting how Oracle systems work by design rather than targeting traditional software bugs. The vulnerability affects Oracle WebLogic Server, a widely deployed enterprise application platform used across financial services, healthcare, retail and government sectors.

This matters for UK businesses because it represents a shift in how attackers approach enterprise infrastructure. Rather than waiting for new vulnerabilities to be disclosed, threat actors are increasingly studying how complex enterprise platforms are designed to operate and finding ways to abuse legitimate functionality. For organisations running Oracle WebLogic Server or other Fusion Middleware components, this is not just about applying a single patch but understanding whether your patching processes can keep pace with Oracle's quarterly critical patch updates, which often contain hundreds of fixes. The three-day deadline CISA has imposed reflects the severity of active exploitation and the speed at which attacks can spread once proof-of-concept code becomes available.

Why it matters

For UK businesses running Oracle infrastructure, this is a prompt to review whether your patching cadence for Oracle products is adequate and whether you have visibility of all Oracle components in your environment. Many organisations struggle with Oracle's complex licensing and deployment models, which can make it difficult to identify where WebLogic Server or other middleware components are running, particularly in development, testing or legacy environments that may not receive the same patching discipline as production systems.

Source: The Register

AI models vulnerable to poisoning through malicious web pages

The Hacker News reports that security researchers at Oasis Security have disclosed a vulnerability in NVIDIA NemoClaw that allows attacker-controlled web pages to take unauthenticated control of local Ollama instances serving AI agents and plant hidden instructions directly inside the model itself. The weakness stems from how NemoClaw handles networking when connecting to local AI model servers. An attacker who can get a user to visit a malicious webpage could exploit this to inject persistent instructions into the AI model that would affect all subsequent interactions, effectively poisoning the model's behaviour without the user's knowledge. Separately, Dark Reading reports that researchers have demonstrated how hidden HTML prompts invisible to users can manipulate AI-powered email summarisers into producing false or malicious summaries of email content.

This matters for UK organisations because AI agents and AI-assisted tools are rapidly being integrated into business workflows, from email summarisation and document analysis to customer service and internal knowledge management. Many of these tools run locally or connect to local model servers, and the assumption has been that local deployment provides better security control than cloud-based AI services. These findings challenge that assumption by showing that local AI infrastructure can be compromised through relatively simple web-based attacks. For organisations deploying AI coding assistants, AI-powered productivity tools or autonomous AI agents, the risk is not just data leakage but model corruption that could persist across sessions and affect decision-making, code generation or business logic without obvious signs of compromise.

Why it matters

For many organisations exploring or deploying AI tools, this is a reminder to review how AI agents and local models are networked, what authentication controls are in place, and whether AI-generated content is being validated before it influences business decisions. The risk is particularly relevant for organisations using AI coding assistants, AI-powered email tools or AI agents that interact with internal systems, where poisoned model behaviour could introduce subtle errors or security weaknesses that are difficult to detect.

Source: The Hacker News

Sophisticated phishing platform enables real-time session hijacking

Infosecurity Magazine reports that researchers have identified a phishing-as-a-service platform called ZeroTokens that gives attackers live control of victim sessions in real time, targeting 53 financial brands including banks, payment processors and cryptocurrency platforms. Unlike traditional phishing kits that simply harvest credentials, ZeroTokens acts as a transparent proxy between the victim and the legitimate service, allowing the attacker to observe and manipulate the session as it happens. This enables attackers to bypass multi-factor authentication, intercept one-time codes, and complete fraudulent transactions while the victim believes they are interacting with the genuine service. The platform is being offered as a subscription service to other criminals, lowering the technical barrier for conducting sophisticated phishing attacks.

This matters for UK businesses because it represents the industrialisation of advanced phishing techniques that were previously only seen in targeted attacks. The real-time session hijacking capability means that even organisations with strong MFA controls can be bypassed if users are tricked into interacting with a ZeroTokens phishing page. For financial services, fintech companies, payment processors and any organisation handling financial transactions, this threat is particularly acute because attackers can complete fraudulent transactions during the live session before security teams have any opportunity to intervene. The fact that this capability is now available as a service means the volume and sophistication of phishing attacks targeting financial services is likely to increase significantly.

Why it matters

For UK businesses in financial services or handling payment transactions, this is a prompt to review whether your fraud detection systems can identify suspicious session behaviour in real time, not just at the point of credential entry. It also reinforces the importance of user awareness training that focuses on recognising phishing attempts through URL verification and understanding that MFA alone is not a complete defence against session hijacking attacks.

Source: Infosecurity Magazine

CISA red team assessments reveal detection and response gaps

CISA has published findings from simultaneous red team assessments conducted at two organisations, revealing significant differences in defensive outcomes despite both organisations ultimately suffering full domain compromise. In both environments, the red team achieved complete control of the domain and accessed sensitive business systems and cloud resources. However, Organisation A failed to detect or contain any of the red team's activities throughout the entire assessment period, while Organisation B detected multiple stages of the attack and took containment actions, though not quickly enough to prevent compromise. The assessment found that Organisation A's security operations centre lacked clear processes for triaging alerts, had no defined escalation procedures, and did not conduct proactive threat hunting. Organisation B, by contrast, had established detection and response procedures, conducted regular threat hunting, and maintained clear communication channels between security teams and system owners.

This matters for UK businesses because it demonstrates that having security tools deployed is not the same as having effective detection and response capability. Both organisations in the assessment had security monitoring in place, but only one had the operational discipline, clear ownership and practiced procedures needed to actually use that monitoring to detect and respond to attacks. For many UK organisations, particularly those in sectors facing increased regulatory scrutiny around operational resilience, this highlights the gap between technical security controls and the human processes, team structures and organisational habits that determine whether those controls deliver meaningful protection. The findings are particularly relevant for organisations preparing for NIS2 compliance, where incident detection and response capabilities will be subject to regulatory oversight.

Why it matters

For UK businesses, this is a reminder that mature security comes from clear ownership, practiced procedures and regular testing of detection and response capabilities, not just from deploying security tools. Organisations should review whether their security operations teams have documented processes for alert triage and escalation, whether they conduct proactive threat hunting, and whether they regularly test their ability to detect and respond to realistic attack scenarios through exercises or red team assessments.

Source: CISA

Today's Key Actions

• Review your Oracle WebLogic Server and Fusion Middleware patching status and ensure you have a process for applying Oracle critical patch updates within a reasonable timeframe, particularly for internet-facing systems.
• If your organisation is deploying AI agents, AI coding assistants or AI-powered productivity tools, review how these systems are networked, what authentication controls are in place, and whether AI-generated content is validated before it influences business decisions or code.
• For organisations in financial services or handling payment transactions, review whether your fraud detection systems can identify suspicious session behaviour in real time and ensure user awareness training covers URL verification and the limitations of MFA against session hijacking.
• Assess whether your security operations team has documented processes for alert triage and escalation, conducts proactive threat hunting, and regularly tests detection and response capabilities through realistic exercises.
• Ensure there is clear ownership across the organisation for maintaining security monitoring effectiveness, not just deploying security tools, and that communication channels between security teams and system owners are established and regularly used.

Secarma Insight

Today's stories reinforce that effective security is built on operational discipline, clear ownership and realistic understanding of how attacks actually unfold in your environment. The gap between having security controls and being able to use them effectively is often where organisations are most vulnerable. Mature security practice means understanding that attackers are studying how your systems are designed to work, not just looking for software bugs, and that AI tools and advanced authentication controls create new risks as well as new capabilities. The organisations that respond most effectively to incidents are those that have practiced their response procedures, established clear escalation paths, and built security into their operational habits long before an incident occurs.