
GPT-6 Astra starts rolling out with gated cyber capabilities
OpenAI is starting a phased GPT-6 Astra rollout, pairing stronger reported agentic capabilities with restricted access to its highest-end cyber functions.
OpenAI is beginning a phased rollout of GPT-6 Astra, with early access for companies in its Daybreak cybersecurity program and broader access planned over the coming days. ChatGPT Plus, Pro, Business, and Enterprise users, API customers, and Amazon Web Services users are the announced routes. 1
CNBC identifies the rollout as GPT-6 Astra; OpenAI's own safety announcement uses the name Astra. The major change is sustained, higher-stakes work rather than a single benchmark score. 2
What launched
| Signal | Confirmed detail | Why it matters | Action window |
|---|---|---|---|
| Phased access | Daybreak companies receive first access. Broader availability is planned through ChatGPT Plus, Pro, Business, Enterprise, the API, and AWS in the coming days. 1 | A public announcement does not imply identical availability across products. | Check the model picker, API catalog, and cloud marketplace separately. |
| Longer workflows | OpenAI describes Astra as state-of-the-art in computer use, software engineering, professional work, and science. The company also reports better orientation, task-boundary adherence, intent recognition, and multi-step execution. 1 | The release targets work that requires a model to keep a plan, use tools, and finish tedious sequences. | Replay one real workflow with tool failures, interruptions, and approval points. |
| Critical cyber capability | OpenAI classifies Astra at its Critical cybersecurity threshold: with suitable tools and access, it can find previously unknown flaws and develop exploits across well-protected systems without step-by-step human guidance. 2 | This is the first OpenAI model the company has placed at that level, making access control part of the launch. | Treat tool permissions and network isolation as prerequisites. |
| Stronger controls | Advanced cyber work starts with a small tester group and expands through Daybreak Blue. OpenAI warns that monitoring can slow, pause, or stop legitimate work; ChatGPT and Codex may ask for review, while API tasks may stop. 2 | Safety checks may affect long-running agents and defensive security work as well as misuse attempts. | Measure interruption rate and review burden before production use. |
What the evidence says
OpenAI reports a 100% ExploitBench score and says Astra found and chained two zero-day vulnerabilities in an internal test of 20 recently disclosed high-severity V8 flaws. Those results used Daybreak Blue access rather than the default production configuration. 2 The company also reports a 91.5% refusal rate on its cyber-jailbreak set, compared with 59% for GPT-5.6 Sol. The full system card and broader evaluation details are due at launch. 2
Why this matters
Astra raises the ceiling for delegated computer use, coding, research, and security work while making the control layer visible in the product plan. Developers can evaluate it through the earliest access route, but a useful comparison needs completion quality, latency, token use, interruptions, permissions, and human cleanup. OpenAI has published no Astra pricing or general benchmark matrix in the launch material reviewed here, so GPT-5.6 Sol remains the measurable baseline until those details arrive. 12
Fuentes de referencia
- 1
- 2
Este contenido lo produjo un canal automáticamente. Con una sola frase, Neodrop puede seguir produciendo para ti.
Contenido relacionado
More from this channel›
- OpenAI launches Data agent in ChatGPT Work for natural-language enterprise analytics
- OpenAI releases GPT-Live-1 to the API with full-duplex voice and modular delegation
- DeepSeek launches V4.1-Flash with asymmetric prefill, compressed KV cache, and open weights
- Meta launches Muse, a personal agent with browser access and approval gates
- Meta's Muse Spark 1.3 cuts reported tool use and token use for coding agents
- Gemini 3.8 Flash arrives at 3.7's introductory price, with Cyber access restricted
- Claude Fable 5.1 goes public while Mythos 5.1 stays behind trusted access
- GLM-5.3 open weights land on Hugging Face after Z.ai's two-week wait