OpenAI’s new model leaps ahead in coding capabilities—but raises unprecedented cybersecurity risks

OpenAI’s new model leaps ahead in coding capabilities—but raises unprecedented cybersecurity risks

https://www.aol.com/articles/openai-model-leaps-ahead-coding-211904459.html

Publish Date: 2026-02-25 10:37:00

Source Domain: www.aol.com

• Advanced Coding Performance: OpenAI’s latest model, GPT-5.3-Codex, exhibits superior performance on coding benchmarks compared to previous AI models by both OpenAI and its rivals, signaling a strategic advantage in AI-assisted coding.

• Cybersecurity Concerns: The powerful coding abilities of GPT-5.3-Codex raise significant cybersecurity risks, leading OpenAI to implement stringent controls and delay full developer access to mitigate potential harm.

• Limited Initial Release: The model is initially available only to paid ChatGPT users for routine software development tasks, and full API access for high-risk uses is restricted.

• Safety and Safeguards: To address cybersecurity risks, OpenAI has introduced comprehensive safety measures, including a trusted-access program for vetted security professionals to manage high-risk applications.

• Financial Support for Defense: To encourage the development of cybersecurity defenses, OpenAI is offering $10 million in API credits for developers working on enhancing cyber defenses.

• Preparedness Framework: The release of GPT-5.3-Codex marks the first time a model has been rated a “high” risk in terms of cybersecurity, leading to the implementation of detailed safeguards following OpenAI’s Preparedness Framework.

• New Risk Levels: The framework ensures that no “high” risk models are released without adequate safeguards, emphasizing cybersecurity as a critical focus area.

• Monitoring and Enforcement: Additional protections include safety training, automated monitoring of the model’s behavior, and enforcement pipelines that incorporate threat intelligence to preempt potential abuses.