OpenAI has introduced GPT-5.5-Cyber, a modified version of its flagship AI model designed for cybersecurity tasks. It is currently available to a limited group of developers.

This neural network is not intended to significantly expand capabilities in the digital environment; rather, it has been trained for more flexible handling of security-related tasks, such as identifying and classifying vulnerabilities, verifying patches, and analyzing malware.

“GPT-5.5-Cyber will allow a select group of partners to explore complex workflows where access specificity matters,” the startup stated in its blog.

The release from OpenAI comes a month after the launch of Mythos, an AI model from Anthropic that garnered significant attention for its cybersecurity capabilities. The company opted not to release the neural network publicly due to high security risks. Instead, it introduced Project Glasswing, an initiative to test the tool in secure conditions.

In April, OpenAI released GPT-5.4-Cyber in response, aimed at identifying vulnerabilities as part of the Trusted Access for Cyber program launched in February.

Voice Models

OpenAI did not stop there. The company unveiled three voice models:

  • GPT-Realtime-2 — the first neural network with GPT-5-level logical reasoning capabilities, able to handle more complex queries and engage in natural conversations;
  • GPT-Realtime-Translate — an LLM for real-time translation that keeps pace with the speaker's speech;
  • GPT-Realtime-Whisper — a streaming speech-to-text system that operates in real time.

Introducing GPT-Realtime-2 in the API: our most intelligent voice model yet, bringing GPT-5-class reasoning to voice agents.

Voice agents are now real-time collaborators that can listen, reason, and solve complex problems as conversations unfold.

Now available in the API… pic.twitter.com/2DY1LU2vO8

— OpenAI (@OpenAI) May 7, 2026

GPT-Realtime-2 features five levels of reasoning: minimal, low, medium, high, and very high. The context window has been expanded from 32,000 to 128,000 tokens.

GPT-Realtime-Translate translates live speech from over 70 languages into 13. Its context window is 16,000 tokens.

GPT-Realtime-Whisper can generate subtitles during meetings, conferences, classes, and broadcasts, as well as create notes and summaries during or immediately after conversations.

These models are available in the Realtime API.

Codex Extension for Chrome

OpenAI has released a Codex extension for the Chrome browser on macOS and Windows. This tool interacts with applications and websites, working in parallel across all tabs in the background.

Codex now works directly in Chrome on macOS and Windows.

It’s even better at working with apps and sites in Chrome, and now works in parallel across tabs in the background without taking over your browser.

To get started, install the Chrome plugin in the Codex app. pic.twitter.com/pjtHd9gC69

— OpenAI (@OpenAI) May 7, 2026

“Codex can quickly handle routine tasks in the browser, such as navigating structured pages and executing complex data entry processes,” the announcement stated.

At each step of task execution, the model autonomously selects the most suitable tool.

Emergency Contact in ChatGPT

OpenAI has added an “Emergency Contact” feature to ChatGPT. This allows users (or parents in a child's account) to designate a contact person for emergencies.

If the company’s “trained specialists” detect that a user has asked questions about self-harm, a notification will be sent to the designated contact. Users can choose the method of communication—messenger, email, or another ChatGPT account.

As a reminder, in September 2025, OpenAI implemented parental controls in the chatbot.