We’re Tracking the Future of Security

Category: News

Current, event-driven reporting, announcements and industry developments.

  • Nvidia Launches AI Agent Safety Platform With Hardware-Based Watchdog

    Nvidia Launches AI Agent Safety Platform With Hardware-Based Watchdog

    Nvidia has announced the Open Agent Safety Platform, a combination of open-source runtime controls and a reference system design intended to keep autonomous AI agents within defined operational boundaries. The platform separates policy enforcement from the agent itself and is designed for use from testing through production deployment.

    OpenShell applies runtime policy outside the agent

    The software component, OpenShell, places agents in managed sandboxes and routes outbound requests through a supervisor. Policies can restrict file, process, network and API activity. Agents may propose policy changes, but they cannot approve their own requests. Real API credentials are substituted outside the agent workload only for authorized destinations.

    Nvidia also describes Sentry, an optional watchdog running on BlueField-4 data processing units. Because the monitor operates separately from the host, it can observe activity and enforce policy even if the host environment is compromised. Nvidia says the design can quarantine an agent that crosses its boundary.

    Independent enforcement improves auditability

    Organizations evaluating agent platforms should test fail-closed behavior, credential isolation, policy-change approval and the completeness of audit logs. Hardware separation can add resilience, but it does not replace least privilege or application-level validation. SectechMedia tracks these issues in its cybersecurity coverage.

    Sources

  • DC Health Agency Says Hidden Report Data Exposed Nearly 400,000 Beneficiary Records

    DC Health Agency Says Hidden Report Data Exposed Nearly 400,000 Beneficiary Records

    The District of Columbia Department of Health Care Finance is notifying nearly 400,000 Medicaid and DC Healthcare Alliance beneficiaries after discovering that two reports on its public website contained hidden personal information. The incident was not described as a network intrusion; the exposure came from underlying report data that unauthorized users could potentially reach.

    Summary reports contained accessible supporting records

    The reports were intended to display enrollment totals and other aggregate statistics. DHCF said the visible pages did not show personal details, but supporting information may have been accessible between 2023 and July 2026. The affected population includes beneficiaries enrolled during that period.

    The exposed fields varied by record and included Medicaid identifiers and other personal information. DHCF removed the reports, investigated the issue and began notifying affected people. The agency’s notice provides current details and assistance information.

    Publication workflows need hidden-data testing

    Organizations publishing dashboards, spreadsheets or generated reports should inspect source files, embedded datasets, metadata and export endpoints before release. Access tests must verify what an unauthenticated user can retrieve, not only what the page displays. SectechMedia follows related controls in its cybersecurity coverage.

    Sources

  • Former US Soldier Sentenced to 70 Months for Hacking and Extortion Scheme

    Former US Soldier Sentenced to 70 Months for Hacking and Extortion Scheme

    A former US Army soldier has been sentenced to 70 months in prison for participating in a hacking and extortion scheme that targeted at least ten US technology and telecommunications companies. The US Department of Justice said Cameron John Wagenius and accomplices stole credentials, accessed corporate systems and threatened to publish or sell stolen data.

    Stolen access supported extortion and fraud

    According to court records, the group used credentials obtained from victim networks and coordinated through online channels. Prosecutors said the conspirators attempted to extort at least $1 million from data owners, advertised stolen information on cybercrime forums and used some records for additional fraud, including SIM swapping.

    Wagenius had previously pleaded guilty to charges including conspiracy to commit wire fraud, aggravated identity theft and extortion related to computer fraud. The sentence also included $294,978 in restitution. The case covers conduct from April 2023 through December 2024.

    Telecom access demands layered controls

    Telecommunications providers should apply phishing-resistant authentication, isolate administrative interfaces, monitor credential reuse and require independent approval for sensitive account changes. Incident evidence should connect access logs, customer-record queries and extortion communications. SectechMedia follows these controls in its cybersecurity coverage.

    Sources

  • New Mexico Jury Finds Facebook Liable for Misleading Users About Privacy Protections

    New Mexico Jury Finds Facebook Liable for Misleading Users About Privacy Protections

    A New Mexico jury has found Facebook liable for misleading users about privacy protections under the state’s consumer protection law. The verdict followed a two-week trial focused on statements about user data safeguards and the company’s oversight of third-party applications. The judge will determine penalties separately.

    The verdict covers more than 43 million violations

    New Mexico’s Department of Justice said jurors found more than 43 million violations. The state argued that Facebook misrepresented protections around data collected through third-party applications, including conduct associated with the Cambridge Analytica scandal. The jury also found that the company misled the public about reviews of outside developers that accessed user information.

    Meta disputed the verdict and said it would continue to defend its privacy record. Jurors did not accept every state allegation, and the final financial consequence remains unresolved until the penalty phase. Reporting should therefore distinguish the liability finding from any future award.

    Privacy claims need verifiable controls

    Digital platforms should be able to demonstrate how third-party access is reviewed, how violations are detected and how public privacy statements map to operational controls. Audit evidence should cover data flows, developer permissions and remediation. SectechMedia tracks related issues in its cybersecurity coverage.

    Sources

  • JADEPUFFER Attackers Used Compromised Service Principals to Delete Azure Resources

    JADEPUFFER Attackers Used Compromised Service Principals to Delete Azure Resources

    Microsoft has documented a destructive Azure intrusion linked to the threat actor it tracks as Storm-3168, also known as JADEPUFFER. The attackers used two compromised service principals inside one tenant to enumerate resources, collect credentials and attempt to delete cloud services. Microsoft said the activity occurred in early June 2026 and unfolded over roughly 18 hours.

    Cloud identities enabled broad destructive actions

    One service principal performed prolonged reconnaissance while another carried out more than 150 destructive or credential-related operations in about 35 minutes. Targets included storage accounts, virtual machines, Key Vault, Function Apps, App Services and Azure SQL databases. Most targeted storage accounts were deleted, while resource locks and deletion protection blocked some attempts.

    Microsoft found that credentials for one service principal had previously been exposed in a public GitHub issue. Although the secret was later removed, it remained visible in the issue history. The company assessed the operation as ransomware-aligned, but reported no ransom note or confirmed data exfiltration.

    Independent recovery controls limited damage

    Organizations should rotate exposed application secrets, restrict service-principal privileges and alert on unusual resource enumeration or deletion. Recovery safeguards should be administratively independent from identities that manage production resources. SectechMedia follows related risks in its cybersecurity coverage.

    Sources

  • File Notification APIs Can Leak User Activity Across Windows, Linux and Android

    File Notification APIs Can Leak User Activity Across Windows, Linux and Android

    Researchers at Graz University of Technology have shown that operating-system file notification features can act as side channels for user and application activity. The mechanisms are designed to alert software when files change, but event timing and path information can reveal behavior even when an attacker cannot read file contents.

    The exposure varies by operating system

    The demonstrations included keystroke timing and website fingerprinting on Linux, cross-user path visibility on Windows and WhatsApp media events on Android. Most scenarios require an attacker to run code locally, often under another account; the Android test used an application requesting no permissions. The work does not show remote compromise by itself, and the researchers said they were unaware of exploitation in the wild.

    Linux has partially hardened device-file events associated with one of the more serious paths. Microsoft told the researchers that the Windows behavior is by design and emphasized that it exposes paths rather than file contents. The project lists no general fix for several other scenarios.

    Local isolation still matters

    Administrators should limit untrusted local code, separate sensitive workloads and monitor unusual use of file-watching APIs where telemetry is available. Application developers should minimize sensitive information in file names and temporary paths. SectechMedia follows endpoint risks in its cybersecurity coverage.

    Sources

  • SalesBleed Flaws Enabled Zero-Click Data Exfiltration From Salesforce Agentforce

    SalesBleed Flaws Enabled Zero-Click Data Exfiltration From Salesforce Agentforce

    Researchers disclosed three vulnerabilities, collectively named SalesBleed, that could turn untrusted customer-relationship records into instructions for Salesforce Agentforce agents. Two attack paths enabled zero-click data exfiltration, while a third could use an Agentforce-Slack integration to distribute phishing messages under a trusted agent identity.

    Poisoned lead data became agent instructions

    The research used Salesforce Web-to-Lead forms to place hidden instructions inside CRM records. Those instructions remained dormant until an employee asked an Agentforce agent to process the record. Researchers found ways to bypass Trusted URLs controls and embed sensitive CRM information in outbound requests. A related path used Slack link previews and agent messaging behavior.

    Zenity reported the issues to Salesforce in June. Salesforce said the bugs were addressed by August 19, strengthened relevant controls and changed default behavior for certain Slack actions to require user confirmation. The company said it had no evidence of customer exploitation.

    Agent permissions require data-flow testing

    Organizations should treat external CRM fields as untrusted input, restrict agent-accessible data, require confirmation for high-impact actions and monitor outbound destinations. Testing should verify actual data flows rather than assume an allowlist blocks every parsing variation. SectechMedia follows these issues in its cybersecurity coverage.

    Sources

  • CISA Election Security Plan Highlights Patching and Voter-Database Risks

    CISA Election Security Plan Highlights Patching and Voter-Database Risks

    The Cybersecurity and Infrastructure Security Agency has published its 2026 election security plan, outlining operational risks and federal support priorities for state and local election organizations. The plan treats elections as a distributed critical-infrastructure environment in which technology, physical processes and public communications must remain resilient together.

    Patching and voter databases remain central concerns

    CISA notes that election offices can face practical barriers to patching systems during active election periods, when changes may disrupt testing or certified configurations. The plan also emphasizes risks to voter-registration databases, including unauthorized access, service disruption and manipulation attempts. Recommended support includes vulnerability information, incident coordination, exercises and resources tailored to election administrators.

    The document also addresses physical security, continuity planning and the need to communicate quickly when technical incidents or false claims could undermine confidence. Responsibility remains shared among local, state and federal organizations rather than concentrated in one platform.

    Controls need evidence before election day

    Election authorities should maintain tested offline recovery, privileged-access review, logging and clear change windows. Tabletop exercises should connect cyber teams with physical-security, communications and legal functions. SectechMedia tracks this combined risk in its cyber-physical security coverage.

    Sources

  • Microsoft Pauses KB5002907 After Reports of Office License Deactivation

    Microsoft Pauses KB5002907 After Reports of Office License Deactivation

    Microsoft has paused distribution of optional update KB5002907 after administrators reported that it deactivated valid perpetual Microsoft Office licenses. The update was intended for out-of-date Microsoft 365 Apps installations, but reports described Office 2016 and Office 2019 systems entering an unlicensed state after installation.

    The update was withdrawn while Microsoft investigates

    Microsoft’s support entry identifies KB5002907 as an optional servicing update for older Microsoft 365 Apps installations. According to incident reporting and administrator feedback, affected devices could lose activation even when a legitimate perpetual license was present. Microsoft stopped offering the update while it reviewed the behavior.

    Organizations that did not receive the update do not need to remove anything. Administrators managing affected systems should consult Microsoft’s current support guidance before attempting repair, because licensing state and deployment channel can change the appropriate recovery action.

    Update validation should include business functions

    Patch testing often focuses on boot, application launch and security telemetry. This incident shows why organizations should also validate licensing, authentication and line-of-business workflows before broad deployment. A staged ring, rollback plan and inventory of update channels reduce the operational impact of defective releases. SectechMedia covers related resilience issues in its cybersecurity coverage.

    Sources

  • IRS Considers AI Triage for Identity-Theft Backlog Nearing 257,000 Cases

    IRS Considers AI Triage for Identity-Theft Backlog Nearing 257,000 Cases

    The US Internal Revenue Service is considering whether artificial intelligence could help assess the complexity of identity-theft cases and route them to appropriately trained employees. The proposal appears in the agency’s response to a Treasury Inspector General for Tax Administration audit; the report does not say that an AI triage system has already been deployed.

    Most delay occurs before assignment

    The audit found that identity-theft victims waited an average of roughly 20 months for resolution across fiscal years 2023 through 2025, compared with the agency’s 120-day objective. Cases in the audit sample spent most of that time in unassigned inventory. The unresolved workload had fallen from earlier levels but still stood near 257,000 cases in May 2026.

    TIGTA recommended evaluating case complexity before assignment and moving claims into active processing more quickly. The IRS agreed to study whether AI could support that work rather than divert experienced identity-theft staff to screen incoming cases.

    Automation needs auditable safeguards

    Any triage model used for identity-fraud cases would need documented routing criteria, human review, bias testing and an appeal path for misclassified claims. Performance should be measured against resolution time and error rates, not only throughput. SectechMedia follows related controls in its identity and access coverage.

    Sources