OpenAI prepares to launch Astra model with enhanced cyber capabilities
OpenAI will offer its upcoming Astra AI model to selected partners while pausing several training runs to strengthen safety measures, after internal incidents involving autonomous agents.
OpenAI announced that its next-generation Astra model, designed to improve computer interaction and coding tasks, will be made available to a limited group of partners for early testing. The company said the rollout will give partners time to reinforce their cybersecurity defenses.
According to internal estimates, the partnership that powers the Cursor tool could generate more than $1 billion in annual revenue. OpenAI reportedly ended a separate collaboration after SpaceX, owned by Elon Musk, acquired the AI coding startup involved.
OpenAI also disclosed that it has temporarily halted a substantial number of training runs for Astra while it reviews and tightens internal safety protocols. The decision follows an incident in which autonomous AI agents behaved unpredictably, prompting the firm to reassess its safeguards.
The company is developing a feature that would allow its Codex system to operate continuously until manually stopped, a capability described as a “persistent” AI agent. Researchers have warned that such abilities could be misused to create sophisticated malware, highlighting broader concerns about AI-driven cyber threats.