TLDR
- OpenAI stated its upcoming AI models carry “high” cybersecurity risks and could be used to create zero-day exploits against protected systems
- The company’s GPT-5.1-Codex-Max model reached 76% accuracy on cybersecurity challenges in November 2025, up from 27% three months earlier
- OpenAI is launching multiple defense initiatives including access controls, infrastructure hardening, and enhanced monitoring capabilities
- The company introduced Aardvark, an AI security tool that scans code for vulnerabilities and recommends patches to developers
- A new Frontier Risk Council will bring external cybersecurity professionals to work alongside OpenAI’s internal security teams
OpenAI published a warning on December 10 about cybersecurity dangers posed by its next-generation artificial intelligence models. The company said these advanced systems might develop functional zero-day remote exploits against well-protected computer systems.
The ChatGPT developer explained that upcoming models could assist with complex enterprise or industrial intrusion operations designed to produce real-world effects. OpenAI disclosed this information through a blog post addressing the expanding capabilities of its AI technology.
This announcement adds to growing industry awareness about AI security challenges. Other leading technology companies have also moved to protect their AI systems from potential threats.
Google revealed updates to Chrome browser security earlier this week aimed at preventing indirect prompt injection attacks on AI agents. These security improvements arrived before Google plans to expand Gemini agentic capabilities across Chrome.
In November 2025, Anthropic reported that threat actors, possibly connected to a Chinese state-sponsored organization, had exploited its Claude Code tool for an AI-powered espionage campaign. Anthropic successfully interrupted the operation.
Dramatic Growth in AI Hacking Capabilities
OpenAI provided specific performance metrics demonstrating how quickly AI cybersecurity skills are developing. The company’s GPT-5.1-Codex-Max model scored 76% on capture-the-flag cybersecurity challenges in November 2025.
This marks a substantial increase from GPT-5’s 27% score in August 2024. Capture-the-flag competitions test how effectively systems can identify and exploit security vulnerabilities.
The threefold improvement in just three months illustrates the rapid pace at which AI models are acquiring sophisticated cybersecurity knowledge. These capabilities have applications for both offensive and defensive security operations.
Multiple Defense Layers Being Implemented
OpenAI announced investments in strengthening its models for defensive cybersecurity applications. The company is creating new tools designed to help security professionals audit code and repair vulnerabilities with greater efficiency.
The Microsoft-backed organization is deploying multiple security measures including access controls, infrastructure hardening, egress controls, and comprehensive monitoring systems. OpenAI is teaching its AI models to decline harmful requests while maintaining usefulness for educational and defensive purposes.
The company is improving monitoring across all products that utilize frontier models to identify potentially malicious cyber activity. OpenAI is collaborating with specialized red teaming organizations to evaluate and enhance its safety protocols.
Security Agent and Expert Advisory Group
OpenAI unveiled Aardvark, an AI agent functioning as a security researcher. Currently available in private beta, Aardvark examines codebases for security weaknesses and proposes fixes.
Developers can rapidly implement the patches Aardvark suggests. OpenAI intends to provide Aardvark at no cost to selected non-commercial open source repositories.
The company will introduce a program offering qualified cyberdefense users and customers tiered access to enhanced security capabilities. OpenAI is establishing the Frontier Risk Council, which will unite external cyber defenders and security practitioners with its internal teams.
The council will initially concentrate on cybersecurity issues before expanding into additional frontier capability domains. OpenAI stated it will soon share more information about the trusted access program for users and developers focused on cyberdefense work.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
