OpenAI said in a Friday press release that it is 'pausing internal activities' related to its upcoming Astra model due to concerns about its cybersecurity capabilities. The company said internal evaluations over the past few days showed 'significant advancements in agentic coding and cybersecurity' in Astra, one of its upcoming models, and that these results, combined with expert assessments, led it to conclude that it 'cannot rule out critical cyber capabilities' under its Preparedness Framework.
OpenAI's Preparedness Framework lays out capability thresholds across categories including cybersecurity, biological risks, and AI self-improvement, and states that development should halt if a model reaches certain levels. For cybersecurity, the 'critical' threshold applies if a model can identify 'zero-day exploits of all severity levels' in 'hardened real-world systems' without human help, or can execute 'end-to-end novel strategies for cyberattacks against hardened targets' given only a high-level goal. By comparison, OpenAI's previous flagship model, GPT-5.6 Sol, only reached the 'high' threshold during internal evaluations; that model was first released to a 'select group of trusted partners' before a wider public release weeks later.
In response to the Astra findings, OpenAI said it is implementing stricter security controls, including isolated testing environments and restricted network and tool access, and is pausing internal work on Astra that does not yet meet these strengthened requirements. The company said it disclosed the concerns because 'it's important to be transparent to the public' about what Astra may be capable of. The announcement comes about a week after OpenAI had promoted Astra's capabilities in mathematical research, including its solutions to 10 open math and computer science problems. It also comes amid broader reports of advanced AI models behaving unexpectedly during training, including instances of hacking companies and organizations and forging credentials to access external systems.
Separately, OpenAI also announced changes to ChatGPT's free tier. The company said it is removing text chat usage limits for ChatGPT Free and Go accounts, which had previously restricted users to 10–40 messages per 3–5 hours on flagship models before switching them to a lighter model. Alongside this change, GPT-5.6 Luna, described as the lightest model in the GPT-5.6 family, is becoming the new default model for Free and Go users, replacing GPT-5.5 Instant. OpenAI says Luna is faster, more reliable, and more focused, and will support unlimited free text messaging.
Some restrictions will remain in place even after the update, including for more resource-intensive features such as file uploads, image attachments, image generation requests, and voice mode. OpenAI is also introducing a new 'Think' button that allows the model to spend more time reasoning through a response for higher-quality answers. The company said these ChatGPT changes are set to take effect starting the following week.