Archive · 11 Dec 2025
OpenAI Warns of High Cybersecurity Risks from New AI Models
OpenAI, the company behind ChatGPT, said that their new AI models could be dangerous for cybersecurity.
These models might be able to hack into well-protected systems or help with big hacking operations.
To fight this, OpenAI is making tools to help defend against these threats.
They are also using special security measures to keep their systems safe.
This is important because as AI gets smarter, it can be used in both good and bad ways, and OpenAI wants to make sure they are ready to handle the risks.
OpenAI is not the only company working on this; others like Google are also upgrading their security to protect against AI-led threats.
OpenAI's AI model, GPT-5.1-Codex-Max, recently scored 76% on a test that measures its ability to find and fix security problems, which is a big improvement from earlier versions.
OpenAI is also working with experts to make sure their AI models are safe and helpful.
OpenAI warned that its upcoming AI models could pose high cybersecurity risks if misused.
The models might be used to deploy zero-day exploits or compromise complex enterprise operations.
OpenAI is investing in defensive cybersecurity tools and workflows to mitigate these risks.
Countermeasures include access controls, infrastructure hardening, egress controls, and monitoring.
OpenAI's AI model, GPT-5.1-Codex-Max, scored 76% on capture-the-flag (CTF) challenges, up from 27% by GPT-5.
- Who
- OpenAI
- What
- Warned about high cybersecurity risks from new AI models
- When
- Wednesday, December 10
- Where
- Not specified
- Why
- To inform about the potential cybersecurity threats posed by advanced AI models and the steps being taken to mitigate them
AI Advancements
Cybersecurity Risks
AI Capabilities
AI Advancements
OpenAI highlights the advanced capabilities of its AI models, which can perform complex tasks and improve cybersecurity defenses.
Cybersecurity Risks
The same advanced capabilities pose significant risks if misused, such as deploying zero-day exploits and compromising enterprise operations.
Defensive Measures
AI Advancements
OpenAI is investing in defensive cybersecurity tools and workflows to mitigate risks and protect systems.
Cybersecurity Risks
Critics may argue that despite these measures, the potential for misuse remains high, and more needs to be done to ensure safety.
- Organization
- OpenAI
- Date
- Wednesday, December 10
- Risk Level
- High
- Potential Threats
- Zero-day exploits, complex intrusion operations
- Countermeasures
- Defensive cybersecurity tools, access controls, infrastructure hardening, egress controls, monitoring
- AI Model Performance
- GPT-5.1-Codex-Max scored 76% on CTF challenges, up from 27% by GPT-5
- Defensive Tools
- Aardvark, an AI agent for scanning codebases and proposing patches
