A New Threat for Gmail Users

Google announced the patching of a critical vulnerability in Gmail that could have allowed hackers to steal corporate data through artificial intelligence. The attack was named GeminiJack and potentially affected more than 2 billion users of the email service, including corporate accounts and employees of large companies worldwide. Experts note that this is not just an isolated incident — it is a warning signal about a new category of AI-related threats.

How the GeminiJack Attack Worked

The attack mechanism was simultaneously simple and sophisticated. Attackers embedded hidden instructions in ordinary Google Docs, calendar invites, or emails. To activate the malicious code, the user did not need to click a link, download a file, or perform any action. Traditional security systems, including spam filters and antivirus software, did not react — the attack completely bypassed standard security measures.

Representatives of Noma, the company that discovered the vulnerability, explain: “If an attacker can influence what the AI reads, they can influence what the AI does.” In practice, this looked like hackers inserting malicious commands into budget documents. When an employee performed a standard search in the Gemini Enterprise system, for example, “show our budgets,” the AI automatically retrieved the “poisoned” document and executed hidden instructions, passing confidential information to the attackers.

This type of attack is called a “prompt injection” and demonstrates a new class of vulnerabilities specific to AI assistants. If a user does not read every individual document, email, or invitation, malicious prompts can silently enter the AI’s workflow, which interprets them as ordinary commands.

Expert Warnings and Recommendations

The UK National Cyber Security Centre (NCSC) recommends treating AI assistants with the same caution as new employees, rather than as fully trusted tools. If you would not give a person full access to your passwords and financial data, you should not trust an AI with them without active supervision. Essentially, AI becomes a “new level of access,” and companies must take this into account when deploying any automated systems that handle corporate data.

Google has already fixed the specific GeminiJack vulnerability, but experts warn this is just the beginning. Government agencies and researchers believe similar attacks could become much larger, especially as AI tools spread in corporate and consumer products. Any AI update in browsers, documents, or calendars could potentially be exploited by attackers for new attack scenarios.

Previously, Google faced criticism for allegedly training AI on Gmail user data. The main issue is not whether the system learns from this data, but that the AI gains access to the information. Google denies training on user data, but with updates enabled, the AI can see all information available in Gmail, Google Docs, and Calendar. This opens new avenues for potential abuse and makes careful security management essential.

The Future of AI Security and Corporate Preparedness

Noma notes: “This type of attack will not be the last. It reflects a growing class of AI vulnerabilities that organizations need to prepare for right now.” As enterprises integrate AI tools capable of reading corporate email, documents, and calendars, artificial intelligence ceases to be just an assistant — it becomes a full access level to confidential information.

Data security now depends on how carefully users and companies approach the deployment of new AI updates. Every update requires careful monitoring: which functions are enabled, what access is granted, and what potential risks exist. Experts compare the current situation to an endless game of “closing the stable after the AI horses have run away,” emphasizing that modern technologies require continuous monitoring and adaptive protective measures.

Overall, the GeminiJack incident demonstrates not only a technological vulnerability but also the strategic necessity of revising the approach to AI in corporate environments. It serves as a reminder that artificial intelligence, as a tool for automation and productivity, simultaneously becomes a point of heightened risk for data security, and these threats cannot be ignored.