Threat Briefing: November 21, 2025

A string of high-profile incidents has made three cybersecurity themes impossible to ignore: third-party exposure, infrastructure fragility, and fast-evolving threats. Data leaks tied to Salesforce–Gainsight and ServiceNow AI remind us how easily external tools and default configs can open the door to compromise.

Cloudflare’s outage demonstrated how one vendor can trigger widespread operational pain. And North Korea’s new JSON-based malware technique signals that nation-state adversaries aren’t slowing down—they’re getting smarter.

• Salesforce Probes Data Theft Linked to Gainsight Breach – Salesforce is investigating data theft tied to Gainsight-integrated applications after revoking stolen access tokens used for unauthorized data access. The company stressed that its CRM platform wasn’t compromised; the issue stemmed from external Gainsight connections. The incident echoes a similar 2025 Salesloft breach involving token theft. Gainsight confirmed attackers accessed business contact data, and Salesforce has notified and is supporting affected customers. Bleeping Computer
• Internal Configuration Error Triggers Global Cloudflare Outage – Cloudflare suffered a major outage on November 18, 2025, disrupting sites including X, Uber, and ChatGPT. Initially suspected as a DDoS attack, the incident was ultimately traced to an internal configuration change that created an oversized feature file and overwhelmed the system. Users saw widespread “internal server error” messages until services were fully restored later that day. CEO Matthew Prince acknowledged the impact and emphasized strengthened reliability efforts and the need for solid business continuity planning. Dark Reading
• ServiceNow AI Agents Exposed to Prompt Injection Exploitation – Researchers have found that default configurations in ServiceNow’s Now Assist AI allow second-order prompt injection attacks, enabling malicious actors to exploit agent-to-agent interactions for data exfiltration or privilege escalation. AppOmni noted the issue stems from expected default behaviors, not an AI flaw, allowing benign agents to be manipulated into performing harmful tasks. Organizations are urged to adjust configurations, enable supervised execution, and monitor agent activity—underscoring the need for stronger safeguards as AI adoption expands. The Hacker News
• Five Plead Guilty to Helping North Korean IT Worker Scheme – This month, five individuals, including four U.S. nationals, pleaded guilty to aiding North Korean IT workers in obtaining jobs at over 130 companies. Erick Ntekereze Prince, 30, orchestrated the scheme, earning $89,000 by providing “certified” workers using false identities. The group’s actions led to over $900,000 in salaries and generated more than $2.2 million for North Korea, likely supporting its weapons programs. Other guilty parties include Audricus Phagnasay, Alexander Paul Travis, Jason Salazar, and Ukrainian Oleksandr Didenko, who also agreed to forfeit $1.4 million. SecurityWeek
• North Korean Hackers Use JSON Services to Deliver Malware – North Korean actors have updated the Contagious Interview campaign, using JSON storage services like JSON Keeper and JSONsilo to deliver malware. NVISO researchers report they lure targets on LinkedIn to download trojanized GitHub or Bitbucket projects. The BeaverTail malware harvests data and installs a Python backdoor, InvisibleFerret, while an additional payload, TsunamiKit, enables system fingerprinting and data collection. The campaign reflects the group’s ongoing efforts to hide malicious activity within legitimate developer traffic. The Hacker News