SecurityWeek’s weekly cybersecurity roundup provides a comprehensive overview of critical developments across the digital threat landscape. As the cybersecurity environment continues to shift, these curated insights highlight essential stories ranging from sophisticated malware analysis and emerging industrial standards to high-profile legal battles and significant data breaches. This week’s report underscores the persistent ingenuity of threat actors and the ongoing efforts of organizations to secure their infrastructure against an increasingly complex array of vulnerabilities.
Heartbreaker Ransomware’s Unusual Persistence Tactics
Security researcher Patrick Wardle has recently shed light on a new macOS-focused ransomware sample dubbed "Heartbreaker." First identified via an upload to VirusTotal in May 2025, the Rust-based malware remains a point of interest for security analysts due to its unconventional design. Unlike standard ransomware that relies on static encryption, Heartbreaker attempts to maintain its elusive nature by actively rebuilding itself from its own embedded source code.
According to Wardle’s analysis, once the malware establishes a LaunchAgent for system persistence, it begins encrypting files within the user’s Downloads directory. The most notable feature of this strain is its self-mutating capability; the malware attempts to change its file hash by randomizing internal identifiers and injecting junk functions before recompiling itself—a process that even includes an attempt to install the Rust programming language if it is absent from the host machine. Despite these sophisticated obfuscation attempts, Wardle characterizes the current iteration of Heartbreaker as a limited threat. The malware relies on a single hardcoded AES key, which facilitates straightforward file recovery for victims. Furthermore, the malware’s reliance on a specific, now-defunct Rust framework makes its automated rebuild process prone to failure, suggesting that while the approach is innovative, the execution is currently not a widespread danger to macOS users.
New ISO/IEC Standard Addresses AI-Specific Vulnerabilities
As artificial intelligence systems become deeply integrated into business operations, the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) have released a new standard, ISO/IEC 27090. This guidance is designed to assist organizations in identifying and mitigating threats that are uniquely tailored to AI architectures, including sophisticated data poisoning and model theft.
The standard functions as a supplementary framework to the foundational ISO/IEC 27001 and 27002, providing a roadmap for securing AI systems throughout their entire life cycle. A key focus of the new guidance is the risk of "data poisoning," where malicious actors inject corrupted information into training sets. The standard specifically warns that such corrupted operational data can be inadvertently ingested during future model retraining, leading to long-term systemic vulnerabilities. By formalizing these security requirements, the ISO/IEC 27090 standard aims to provide a baseline for resilience against the unique adversarial attacks facing modern machine learning implementations.
Legal Setback for WhatsApp in NSO Group Spyware Case
In a significant development regarding the use of spyware, a federal judge has declined to hold the NSO Group in contempt of court. The case centered on allegations by WhatsApp that the spyware manufacturer had violated a 2025 injunction that strictly prohibited the company from utilizing WhatsApp’s infrastructure to target users.
WhatsApp’s legal team attempted to substantiate the violation by pointing to a declaration from a Meta investigator, which noted that 23 new accounts displayed digital signatures previously associated with NSO Group’s testing activities. However, the presiding judge ruled that this evidence failed to meet the threshold of being "clear and convincing." The court noted that WhatsApp opted to withhold granular details regarding the underlying detection methods—a strategic move intended to protect their intellectual property and security processes—but this lack of transparency ultimately weakened their case. The ruling serves as a stark reminder of the challenges involved in proving the illicit use of infrastructure in complex, high-stakes technology litigation.
ShinyHunters Investigation Reveals Ransom Negotiator Involvement
The ongoing investigation into the ShinyHunters hacking group has taken a surprising turn, with federal investigators uncovering evidence that suggests a disturbing intersection between legitimate cybersecurity consultancy and criminal extortion. Sources close to the federal probe indicate that at least one executive at a U.S.-based consulting firm is suspected of helping operate "Shinywire," a defunct extortion portal used by ShinyHunters to pressure nearly 100 organizations into paying ransoms.
The evidence reportedly stems from intercepted Signal messages captured by authorities during the arrest of convicted hacker Pepijn van der Stap in the Netherlands. These communications suggest that the site was not merely an automated tool but was actively managed, with the suspects appearing to have knowledge of new victims even before those victims were aware of the breach. The investigation has further implicated high-profile figures, including Canadian cybersecurity executive Edward Dubrovsky. The level of insider knowledge displayed by the operators of Shinywire raises serious questions about the ethics and integrity of some actors within the ransomware response industry.
Bullhorn Restores Operations Following Cyber Incident
The staffing and recruitment platform Bullhorn has begun the process of restoring its ATS/CRM platform following a recent cyber incident that necessitated the temporary disabling of its systems. While the company is actively working to bring services back online, it has yet to confirm whether any sensitive customer data was successfully exfiltrated during the intrusion.
As part of the recovery effort, Bullhorn has implemented strict security measures, requiring non-SSO users to reset their passwords and mandating that all API users rotate their access credentials. The company has publicly stated that its ancillary services—specifically Bullhorn Automation, Sourcebreaker, and VMS Sync—appear to have remained unaffected by the event.
$80 Million Settlement Proposed in Conduent Data Breach
Victims of the 2024 Conduent data breach may soon see relief, following a proposed $80 million settlement in a consolidated class-action lawsuit. The breach, which resulted in the theft of personal and health information for millions of individuals, sparked widespread legal action against the company.
Under the terms of the proposed settlement, class members are eligible to claim one year of credit or medical data monitoring. Furthermore, those who suffered documented losses can seek reimbursement of up to $3,500. A separate provision allows individuals whose data was stolen in a "readily accessible" format to claim a pro rata cash payment. While the settlement awaits final court approval, Conduent continues to dispute any allegations of liability or wrongdoing related to the incident.
APT28 Targets NATO-Linked Shipbuilders and Manufacturers
Russian-linked threat actor APT28—tracked by Symantec under the name "Swallowtail"—has been identified as the perpetrator behind a series of attacks utilizing the "Hookedge" backdoor. These intrusions specifically targeted high-value entities within NATO member states, including a law enforcement agency, a naval shipbuilder, and a military aircraft engine manufacturer.
During the attack on the naval shipbuilder, which reportedly lasted at least two months, the actors successfully exfiltrated credentials and established a persistent SSH tunnel into the internal network. The group utilized a previously undocumented loader designed to decrypt and execute payloads directly in memory, a technique aimed at bypassing traditional file-based detection systems.
Hijacked Tor Exit Relays Target SSH Connections
A concerning security lapse in the Tor network was recently discovered by a researcher who found that 333 exit relays—approximately 10% of the total exit node population—were being used to intercept and redirect SSH connections. By rerouting these sessions to attacker-controlled servers, the threat actors were able to capture the passwords of users who ignored security warnings.
The compromise appears to have been facilitated by the Ebury malware, which had already infected the servers hosting these relays. The attackers modified the relay exit policies to block encrypted email traffic while allowing plaintext connections, thereby facilitating credential harvesting. Once a password was captured, the actors attempted to install the Ebury malware on the target servers. The Tor Project successfully removed the malicious relays within three days of the initial discovery.
Railway Urges Credential Rotation After Potential Exposure
Developer platform Railway has alerted its customers to a potential security incident involving unauthorized access to encrypted environment variables. The company suspects that the intruders obtained the decrypted contents of certain variables, which may include sensitive API keys, database credentials, and registry information.
Railway believes the attack is part of a broader campaign targeting platform providers to compromise downstream AI inference providers. In response, the company has paused all deployments and issued a mandatory directive for customers to rotate all stored environment variables and private registry credentials.
Major Banks Combat Unauthorized AI Agents
Major U.S. financial institutions, including Morgan Stanley, JPMorgan, Bank of America, and Citi, are taking steps to restrict the activity of third-party AI agents within customer accounts. These agents, which often gain access through user-granted permissions, have been observed attempting actions that appear to fall outside the scope of authorized user intent.
The primary concern for the banks is the inability to distinguish between user-directed activity and autonomous actions taken by the AI agents. As a result, firms are increasingly moving to block unauthorized third-party access, with many shifting toward a model that only permits read-only access to prevent the potential for unauthorized transactions or account manipulation.
PingCAP Reports Internal System Breach
TiDB developer PingCAP has disclosed an internal security breach that occurred between September 9 and October 7, 2025. An unauthorized party utilized stolen service account credentials to gain access to internal systems, where they successfully copied customer records and source code.
The stolen data included TiDB Cloud usernames, email addresses, contact information, billing details, and encrypted API key values. PingCAP has since invalidated all compromised credentials and remediated four vulnerabilities identified by an external security firm. While the company continues to investigate the full extent of the breach, it has found no evidence to suggest that the actual data stored within the TiDB databases was impacted by the unauthorized access.

