By Asro 
SecurityWeek’s weekly cybersecurity roundup provides a critical synthesis of the week’s most significant developments. As the threat landscape shifts rapidly, this digest offers a concise overview of events that, while perhaps not dominating the headlines individually, collectively illustrate the complexities facing modern security professionals, from evolving ransomware techniques and supply chain vulnerabilities to the emerging challenges posed by the integration of artificial intelligence in financial and cloud sectors.
Heartbreaker macOS Ransomware and the Persistence of Rust-Based Threats
A new analysis from security researcher Patrick Wardle has brought attention to "Heartbreaker," a Rust-based macOS ransomware strain that highlights the increasingly experimental nature of cross-platform malware. First uploaded to VirusTotal in May 2025, the malware remained undetected by every major antivirus engine for a significant period. Heartbreaker employs a rudimentary persistence mechanism, attempting to establish a LaunchAgent to ensure its presence on the system before proceeding to encrypt files within the user’s Downloads folder.
Perhaps most noteworthy is the malware’s self-rebuilding capability. To evade static signature-based detection, the ransomware is designed to recompile itself by accessing its own embedded source code. It randomizes identifiers and injects "junk" functions—non-functional code intended to alter the binary’s hash—before recompiling. It will even attempt to install the Rust programming language environment if it finds the system lacks the necessary tools. However, Wardle suggests that Heartbreaker is currently more of an academic curiosity than a systemic threat. The malware relies on a single, hardcoded AES encryption key, which renders file recovery trivial for victims. Furthermore, its dependency on a specific, now-defunct Rust framework suggests that the current iteration of the ransomware lacks the longevity and sophistication required for a widespread, successful campaign.
ISO/IEC Standard 27090: A New Framework for AI Security
As artificial intelligence becomes deeply embedded in enterprise infrastructure, the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) have released ISO/IEC 27090. This new standard provides comprehensive guidance on detecting and mitigating risks specific to AI systems, including data poisoning and model theft, throughout the system’s entire lifecycle.
The standard is designed to integrate seamlessly with the widely adopted ISO/IEC 27001 and 27002 frameworks, which govern information security management. A primary focus of ISO/IEC 27090 is the threat of "data poisoning," where malicious actors introduce corrupted data into training sets to compromise model integrity. The guidance also explicitly addresses the risks associated with the long-term retraining of models, warning that corrupted operational data could be inadvertently ingested into future versions of an AI system, thereby creating a persistent security vulnerability.
WhatsApp and the NSO Group Legal Standstill
The ongoing legal battle between WhatsApp and the NSO Group has hit a procedural wall. A federal judge has declined to hold the spyware manufacturer in contempt, rejecting WhatsApp’s claims that NSO Group violated a 2025 injunction that barred the company from utilizing WhatsApp’s infrastructure to target users.
WhatsApp’s case hinged on a declaration from a Meta investigator, which alleged that 23 newly identified accounts displayed digital signatures consistent with those previously linked to NSO Group testing activities. However, the court found this evidence insufficient to meet the "clear and convincing" standard required for a contempt finding. The judge noted that the evidentiary gap was, in part, self-inflicted; WhatsApp chose to limit the specificity of the information it presented to the court, aiming to protect its proprietary detection methodologies. This ruling underscores the inherent difficulties in litigating against sophisticated surveillance software providers, where protecting detection tradecraft often comes at the expense of evidentiary weight in court.
ShinyHunters Investigation and the Role of Ransom Negotiators
Federal investigations into the notorious hacking collective ShinyHunters have taken an unexpected turn, placing the professional "ransom negotiation" industry under the microscope. Sources close to the probe suggest that authorities have uncovered evidence linking at least one executive at a U.S.-based consulting firm to the operation of "Shinywire," a portal used to coerce nearly 100 organizations into paying extortion demands.
This investigation gained significant momentum following the arrest of Pepijn van der Stap in the Netherlands. During the investigation, federal authorities accessed Signal communications that reportedly detail frequent interactions between van der Stap, Canadian cybersecurity executive Edward Dubrovsky, and the aforementioned U.S. consulting executive. The messages suggest a disturbing level of coordination, with the negotiators purportedly receiving information regarding new victims even before those victims were aware of their own security breaches. This development raises serious ethical and legal questions regarding the intersection of incident response services and the criminal extortion ecosystem.
Bullhorn and Conduent Face Ongoing Remediation Challenges
Service disruptions and data breaches continue to test the resilience of major platforms. Bullhorn, a provider of ATS/CRM software, recently suffered a cyber incident that necessitated the temporary disabling of its systems. As the company works to restore services, it has implemented a mandatory security posture reset, requiring non-SSO users to change their passwords and mandating that all API users rotate their credentials. While Bullhorn has confirmed the restoration process is underway, they have yet to determine the full extent of the incident, specifically whether any customer data was exfiltrated.
Simultaneously, the fallout from a 2024 data breach at Conduent has culminated in an $80 million proposed class-action settlement. The breach, which exposed the personal and health information of millions of individuals, led to a consolidated lawsuit. Under the proposed settlement terms—which still await final court approval—affected individuals are eligible for one year of credit or medical data monitoring, along with reimbursement of up to $3,500 for documented losses. While Conduent has agreed to the settlement, the company continues to dispute any allegations of wrongdoing or legal liability.
APT28 Targets NATO-Linked Infrastructure
Russian state-sponsored threat actor APT28, tracked by Symantec as "Swallowtail," has been linked to a series of sophisticated attacks targeting critical infrastructure within NATO member states. The campaign centers on the deployment of the "Hookedge" backdoor, which has been used to infiltrate a law enforcement agency, a naval shipbuilder, and a military aircraft engine manufacturer.
In the case of the naval shipbuilder, the intrusion persisted for at least two months. The attackers utilized multiple techniques to harvest credentials, ultimately establishing a persistent SSH tunnel into the internal network. Once inside, they deployed an undocumented loader designed to decrypt and execute payloads entirely in memory, a technique aimed at bypassing traditional disk-based security controls.
SSH Credential Harvesting via Hijacked Tor Exits
A recent security discovery highlights the risks inherent in anonymized routing. A researcher identified that approximately 333 Tor exit relays—roughly 10% of the total network—were actively redirecting SSH connections to attacker-controlled servers. These relays, which were themselves compromised by the "Ebury" malware, were configured to intercept plaintext credentials from users who ignored standard SSH security warnings.
The attackers modified the exit policies of these relays to block encrypted email retrieval, thereby forcing traffic toward plaintext channels. While the Tor Project removed the compromised relays within three days of the report, the incident serves as a stark reminder of the risks of password reuse. One relay operator confirmed that their own compromise was the result of a reused password, a vulnerability that allowed the attackers to co-opt the infrastructure for their malicious campaign.
Railway and PingCAP Disclosures
Developer platform Railway has issued a security disclosure following the unauthorized access of encrypted customer environment variables. The company believes this breach is part of a broader, coordinated campaign targeting platform companies that provide AI inference services. In response, Railway has paused all deployments and is urging customers to rotate all environment variables and private registry credentials as a precautionary measure.
Similarly, TiDB developer PingCAP has reported a security incident involving the theft of service account credentials. Between September 9 and October 7, an unauthorized party accessed internal systems and exfiltrated customer contact information, billing details, and encrypted API keys. PingCAP has since invalidated the compromised credentials and addressed four identified vulnerabilities, though the company is continuing its investigation into the specific scope of the impact on its customer base.
The Emerging Challenge of Unauthorized AI Agents
Major U.S. financial institutions are grappling with the rise of unauthorized third-party AI agents operating within customer accounts. According to recent reports, these agents often utilize legitimate customer credentials to access banking platforms, frequently performing actions that appear to be independent of direct user input.
This trend has put banks in a difficult position. Because these agents operate using valid user permissions, it is often impossible for financial institutions to distinguish between authorized user activity and automated agent behavior. Consequently, major banks including Morgan Stanley, JPMorgan, Bank of America, and Citi are moving to restrict or outright block third-party agent access, with many institutions shifting toward a "read-only" access model to maintain security and mitigate the risks posed by autonomous, automated transactions.
Muslim 
Lina Irawan