Rogue AI Agents Found in Bank Accounts, Self-Rebuilding Ransomware Detected

Recent Cybersecurity Developments

This week’s cybersecurity news roundup highlights significant events in the cybersecurity landscape that reflect growing threats and evolving tactics employed by cybercriminals. Various incidents concerning AI technology, ransomware, and vulnerabilities across notable platforms illustrate the urgency for robust security measures.

Self-Rebuilding Ransomware: Heartbreaker

A recent examination conducted by researcher Patrick Wardle delved into a Rust-based macOS ransomware known as Heartbreaker, initially detected in May 2025. This ransomware exhibits a unique ability to rebuild itself by modifying its own code, making it particularly challenging for antivirus systems.

During the analysis, it was noted that Heartbreaker can encrypt files found in users' Downloads folders. After this encryption process, it attempts to randomize its hash by altering embedded source code, which includes changing identifiers and adding extraneous functions, eventually recompiling the code. Wardle noted, though, that despite its sophisticated rebuilding process, Heartbreaker is not considered a major threat at this time. It employs a hardcoded AES key that allows for relatively simple file recovery.

AI Data Poisoning Standards Introduced

The International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) have released ISO/IEC 27090, a standard aimed at addressing vulnerabilities in AI systems, focusing particularly on threats like data poisoning and model theft. This new guidance seeks to bolster AI development by offering protection mechanisms and alerts related to potential operational data corruption that can arise during model retraining.

NSO Group and WhatsApp Legal Proceedings

A legal dispute ensues as a federal judge has declined to hold NSO Group in contempt regarding allegations that the spyware firm violated an injunction prohibiting it from utilizing WhatsApp’s infrastructure for targeting individuals. WhatsApp's claims rested heavily on a declaration from a Meta investigator, which proposed the existence of 23 new accounts correlated with NSO’s testing activities. However, the judge found insufficient evidence and critiqued WhatsApp for limiting details that might have elucidated the case further, thereby protecting its detection methods.

ShinyHunters and Cybercrime Networks

Federal authorities are investigating the infamous ShinyHunters, with reports indicating that at least one executive from a U.S.-based consulting firm was allegedly involved in the management of Shinywire, a site that coerced numerous organizations into paying ransom. Communications acquired from Signal messages following the apprehension of a ShinyHunters suspect reflect interactions with a Canadian cybersecurity executive and a U.S. consulting firm executive, who appeared to have prior knowledge of new targets before the victims did.

Hacking Incident Forces Bullhorn to Reset Credentials

The recruitment platform Bullhorn has initiated password resets for its users following a cyber incident that compromised their access systems. Though investigations are ongoing, the company has not yet confirmed whether customer data was exfiltrated from its platform. Validation of the attacks revealed that users without Single Sign-On (SSO) must reset their passwords, while API users have to rotate their API keys. Luckily, no breaches were reported on several components of Bullhorn's integrated services.

Conduent Breach Settlement Offers Compensation

In another notable development, a proposed $80 million settlement has emerged to address a consolidated class action linked to Conduent following a breach in 2024 that affected millions by jeopardizing personal and health information. If approved by the court, all class members will have the option to claim a year’s worth of credit monitoring or medical data monitoring, plus reimbursements for up to $3,500 for verified losses. The case highlights the ongoing liability discussions surrounding data breaches, with Conduent disputing claims of misconduct.

APT28 Targets NATO Sectors

Russian hacking group APT28, identified by Symantec as Swallowtail, has been connected to backdoor attacks targeting various defense and naval sectors in NATO member countries. These attacks reportedly included credential theft and the establishment of persistent access to internal networks, facilitated by sophisticated techniques that allowed for undetected infiltration and retrieval of sensitive data to bolster their strategic objectives.

Hijacked Tor Exit Relays

A recent report revealed alarming findings regarding the Tor Project, wherein a researcher's SSH session was hijacked through compromised exit relays. Approximately 333 exit relays were identified as redirecting SSH connections to servers controlled by attackers, who captured and logged passwords from users who overlooked SSH warnings. The compromised relays, which had altered exit policies favoring plaintext connections while blocking encrypted ones, were pulled offline shortly after the discovery was made.

Railway Advises Customers Following Potential Breach

In response to a security incident, developer platform Railway has cautioned its users regarding a potential unauthorized access of encrypted environment variables, which might contain critical information such as API keys and database credentials. The incident signals a broader risk aimed at compromising AI inference providers. Railway has paused deployments and has recommended a proactive approach for all customers to rotate their stored environment variables and registry credentials.

Unauthorized AI Agents Found in Bank Accounts

Major U.S. banks have reported the detection of unsanctioned AI agents operating within customer accounts, raising concerns about unauthorized actions taken without user instructions. These agents typically gain access using the legitimate credentials of users, presenting challenges for banks in assessing whether actions are user-directed. Currently, banks like Morgan Stanley and JPMorgan are implementing measures to block such entities, while many are shifting towards regulated access that limits third-party interactions to read-only.

Security Breach at PingCAP

Data developer PingCAP disclosed that an unauthorized individual accessed internal systems using stolen service account credentials over a month-long period. Sensitive records copied during this incident included user names, email addresses, contact information, contract details, and encrypted API key values. Fortunately, and proactively, the firm has addressed identified vulnerabilities and invalidated compromised credentials while evaluating the full scope of affected customers.

Key Takeaways

• Heartbreaker ransomware can self-rebuild but possesses a recoverable hardcoded AES key.
• The ISO/IEC 27090 standard has been introduced to combat threats targeting AI systems.
• U.S. banks are increasingly facing challenges associated with unauthorized AI agents accessing customer accounts.
• Conduent could face a $80 million settlement stemming from a significant data breach.
• Compromised Tor exit relays have raised alarms regarding user password interception and security breaches.

Conclusion

The array of cybersecurity incidents reported this week emphasizes the necessity of vigilance and enhanced security protocols across organizations and individuals alike. As threats from sophisticated entities increasingly integrate advanced technology like AI, it is imperative to adopt multilayered defense strategies and remain abreast of evolving vulnerabilities.