Google reveals hackers used AI to exploit weakness in two-factor authentication
Google threat intelligence claims to have identified the first known case of cyber attackers using AI to help develop a zero-day exploit. Elsewhere, LLMs are being used to hide malware and create deepfake-driven influence campaigns.
Image by Omer Taha Cetin/Anadolu/Getty.
Listen to this article
Google threat intelligence claims to have identified the first known case of cyber attackers using AI to help develop a zero-day exploit. Elsewhere, LLMs are being used to hide malware and create deepfake-driven influence campaigns.
- Google identified the first known AI-assisted zero-day exploit designed to bypass two-factor authentication on a system administration tool.
- Promptspy Android malware autonomously executes attacks using Google's Gemini models to analyze interfaces and simulate user gestures in real time.
- Cybercriminals now use AI for vulnerability research, exploit development, and creating evasive malware with decoy logic to avoid detection systems.
- Nation-state actors from Russia, China, and Saudi Arabia deploy AI-generated media including voice cloning for hacktivist influence operations worldwide.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
In a report published Monday, GTIG says that criminals are rapidly moving from using AI for phishing campaigns and attack research to embedding it directly into offensive hacking operations.
The report’s headline finding centers on a cyber crime operation involving a zero-day vulnerability in a popular open-source, web-based system administration tool.
“Our analysis of exploits associated with this campaign identified a zero-day vulnerability implemented in a Python script that enables the user to bypass two-factor authentication (2FA) on a popular open-source, web-based system administration tool.”GTIG AI Threat Tracker report, May 11th.
According to GTIG, the exploit enabled two-factor authentication bypass and was intended for “a mass vulnerability exploitation operation" before researchers disrupted the operation.
Google declined to identify either the threat actor or the affected open-source platform, and it does not name the LLM model it believes was used, although it explicitly ruled out its own AI, Gemini and Claude's advanced bug catcher Mythos.
The finding comes as major AI companies are positioning advanced AI models such Anthropic AI's Mythos and Open AI's ChatGPT 5.5 for bug discovery and software auditing. Google notes however, that these capabilities are just as likely to be used by adversaries "as expert-level force multipliers for vulnerability research and exploit development, including for zero-day vulnerabilities."
Why Google thinks the zero-day was AI-driven
The Google report details how the flaw operated: allowing attackers to bypass two-factor authentication because of a weakness in how the system handled trusted logins.
GTIG believes the vulnerability was hard for traditional security tools to spot because the software appeared to function normally, despite containing a serious security flaw.
Have thoughts about this topic? Others do, too. Join them in the discussion.
Researchers think the AI may have helped develop the exploit because advanced models are good at spotting subtle logic flaws that conventional scanners can miss.
GTIG said it concluded with “high confidence” that AI-assisted tools were involved, citing signs including “educational docstrings,” a “hallucinated CVSS score,” and a highly structured Python coding style commonly associated with large language models.
Android backdoor: Promptspy
Separately, Google also highlighted a new wave of autonomous Android malware called Promptspy, an Android backdoor that integrates Google’s Gemini models into live malware operations.
GTIG said the malware represents a shift toward "autonomous attack orchestration” where AI systems independently scan victim environments and generate commands in real time.
Promptspy reportedly uses a module called “GeminiAutomationAgent” to analyze Android interface layouts, calculate screen geometry and execute actions such as “Click” and “Swipe” through simulated gestures.
The malware can also replay authentication gestures, including lock patterns, and PIN activity, to regain device access.
To prevent removal the malware tools overlays invisible touch interceptors on uninstall buttons, making them appear unresponsive.
According to Google, no apps containing Promptspy were found on Google Play.
“Android users now are automatically protected against known versions of this malware by Google Play Protect, which is on by default on Android devices with Google Play Services,” it adds.
AI used on hacktivist influencer campaigns
Elsewhere the report details how cybercriminals are using AI to create stealthier malware capable of “decoy logic” and junk code insertion to disguise malicious behaviour and evade detection.
Beyond cyber intrusions, GTIG said threat actors from Russia, China and Saudi Arabia are increasingly using AI generated media to amplify influence operations.
In one example, researchers linked a pro-Russian “Operation Overload” campaign to suspected AI voice cloning used to impersonate journalists and manipulate news footage.
Unlock more exclusive Cybernews content on YouTube.