AI moves fast. Stay in the know.

A critical ChatGPT vulnerability allowed attackers to silently exfiltrate sensitive user data through a DNS-based covert channel that bypassed all platform guardrails and security warnings. Security researchers discovered the flaw that exploited the fact that DNS queries were treated as harmless infrastructure, creating a blind spot that enabled data theft without triggering any user alerts or consent prompts.

Source: TechRadar

What to know:

• The vulnerability combined prompt injection with DNS abuse to exfiltrate data through domain name queries rather than monitored HTTP or API channels.
• DNS traffic was treated as "harmless infrastructure" by ChatGPT's security systems, creating a blind spot that did not trigger approval dialogs or risk warnings.
• Attackers could initiate the exploit through malicious prompts embedded in emails, PDFs, websites, or even through custom GPTs posing as legitimate tools (such as "personal doctors").
• Users unknowingly shared highly sensitive information, medical conditions, payment slips, contracts, and private conversations, assuming ChatGPT's environment was fully isolated.
• OpenAI deployed a fix towards the end of February 2026, marking the second major vulnerability patched that week after a separate Codex command injection flaw.

Why it matters:
This incident exposes a fundamental assumption gap in AI security: organizations trust that GenAI platforms prevent unauthorized data extraction, but novel attack vectors continue to emerge. The use of DNS, a protocol designed for basic name resolution, as a data exfiltration channel demonstrates that AI guardrails focused on policy and intent can miss infrastructure-level exploits. For mid-sized businesses deploying ChatGPT across teams, this vulnerability underscores the need for continuous monitoring of AI interactions at the network level, real-time anomaly detection across all data transmission protocols, and proactive risk assessment of GenAI tools before sensitive data enters the conversation.

A recent State of AI Risk Management 2026 report by The Purple Book Community reveals concerning gaps in AI security readiness versus the actual risks faced by organizations. The survey highlights the growing issue of “shadow AI” usage and vulnerabilities introduced by AI‑generated code, both of which are creating substantial security and governance blind spots.

Source: BusinessWire

What to know:

• 59% of organizations report or suspect the use of shadow AI, AI systems operating outside of governance controls, posing significant security risks.
• 70% of respondents acknowledge vulnerabilities in AI‑generated code used in production systems, which may introduce new attack vectors and complicate security efforts.
• These findings underscore the critical need for effective AI governance frameworks that can provide real‑time monitoring, prevent unauthorized AI use, and ensure AI code is secure before deployment.
• The report further emphasizes the need for proactive AI risk management to close the gap between perceived AI security readiness and the realities of potential vulnerabilities.

Why it matters:

The rapid adoption of AI technologies by enterprises, coupled with the increase in shadow AI and unvetted AI code, underscores the importance of comprehensive risk assessment and continuous monitoring. Mid‑sized organizations, in particular, are at risk of security breaches and compliance failures if they do not invest in robust AI governance and security frameworks to manage these emerging threats.